#aistuff

posts tagged #aistuff

0

Currently experimenting with exploration policies for deep RL on Super Mario Bros - Agent is able to beat all levels I threw at it - You can watch the AI learn live

preview

publication croisée depuis : lemmy.pierre-couy.fr/post/2152233

I’ve been playing with deep reinforcement learning for a while. I originally started with a simple DQN, added all improvements from the Rainbow paper, and finally changed C51 for a quantile regression (and plan to swap it for an Implicit Quantile Network).

After implementing C51 (which was my first time with distributional RL) I started playing with policies that take advantage of the learned distributions : By independently taking N samples from each action-value distribution, scoring actions by averaging the samples, and picking the greedy action with respect to these scores, I was able to make the agent learn faster than similar agents using only NoisyNets or an epsilon-greedy policy (I’m still using NoisyNet, this is done on top of it). In the limiting cases, N=1 is just Thompson Sampling and N=+Infinity is just a plain greedy policy.

Finding an optimal value for N proved to be a challenge, so I decided to pick a random value for it at the start of each episode (N = 2**rng.uniform(8,12) for a QR-DQN with 32 quantiles/action works well in my experiments), which led to even better results.

I later found out about DLTV which made the agent discover new behaviors, but performed worse than previous experiments overall. Inspired by it, I tried something I did not find in previous works and got the best results out of all my previous experiments :

At each time step, compute an exploration_score as the ratio of “intra-action variance” over “inter-action variance” (rendered latex equation). I then take N/exploration_score samples from each distribution, and pick an action as described above. (more details at the end of this post)

For anyone reading this, I have a few questions :

  1. Are you aware of any previous work I missed that tries similar exploration policies with distributional RL (interpolating between Thompson sampling and the greedy policy)
  2. Most papers I found about learning from multiple exploration policies seem to be in the context of multi-actor parallelization. Is there any novelty in randomizing the policy parameters at the start of each episode, especially in the single-actor case ?
  3. Is any part of what I’m doing worth the time it would take to quantitatively evaluate it ? I’ve been doing it mainly for learning and fun and have only qualitatively evaluated it so far. However, if there’s a chance I can contribute to the field, I’ll gladly make some time to compare it to published papers on ALE.

A few more details

I actually track a moving average and standard deviation of the exploration score, which lets me shift/rescale its values to a target average and standard deviation, and divide N by the shifted/rescaled value. I initially started with a target average of 1 and standard deviation of 1 as well (which gave good results), then tried randomizing these parameters at the start of each episode as well. This led to a lot more diversity in the policies and even better results.

Since this worked so well, I additionally randomized the noise strength in the NoisyNet layers.

Overall, this made the agent a lot more robust to deviating from what it considers to be the optimal trajectory, and allowed it to learn complex behaviors previous iterations were never able to learn (e.g. taking a few steps back to gain momentum, waiting for good cycles, or dodging hammer bros)


Watch it learn

For anyone interested, I made a live stream of the training in progress with graphs and some more details on the experiments I’m running. The current training run was started ~2.5 days ago. The agent has finished and unlocked levels up to 5-1, and is currently learning 5-2.


A lot more details

Long text hidden, click to expand

Available actions : The agent does not have access to the up and down buttons, the available actions only use left, right, A and B. Adding the down button would double the total number of actions (because down can be pressed on top of all available actions). Reward function : It mainly consists of reward(t) = max(0, x(t) - previous_best_x) + a larger reward for beating a stage. I had to tweak the scaling of both components. I initially had penalties for time and death, but one made the agent suicidal in front of hard-to-overcome obstacles, while the other made it fear them too much and hug the left side of the screen. Removing both proved to increase the performance. One trick that seems to help with most ‘*-3’ levels (which have a lot of void to fall into) was to hold the reward while the vertical velocity of Mario is negative (meaning it is falling). Without this trick, the agent would sometimes get stuck learning to jump the farthest it can into the void. Stage scheduling : Each episode is one attempt on one level. At the start of each episode, a stage is randomly picked with probability proportional to 1/(number of times the stage was beaten) among the unlocked stages. Each stage is unlocked after the previous one has been beaten 30 times, with only 1-1 unlocked at the start of the training. Available stages : The first iterations of the agent were unable to learn maze castles (4-3, 7-3 and 8-4), so I removed them all. The reward function will give rewards for the first path the agent tries, then the agent will be teleported back by the game and no reward is received until it finds the right path and gets past the point where the game teleported it back. I plan to test newer (better) versions of the agent on these stages only and see if mazes can be re-added to the pool. I’ve also removed underwater stages (2-2 and 7-2). The agent can learn them fine, but the game dynamics are really different from all other stages and they’re really boring to watch. Since I already removed a bunch of stages, I figured I could remove these as well but I may re-add them with mazes because beating every level is cooler than beating a cherry-picked selection. Since 8-4 is the only stage that requires going down a pipe, I considered it was not worth it to add the down action and will likely never re-add it to the pool, which would unfortunately be really anti-climactic… Replay buffer warm-up : After initially using the standard approach of filling the buffer with transitions sampled from a random policy before training the neural net, I came-up with a “soft warm-up” scheme in which the first gradient updates happen after only 2000 transitions, but initially happen every few thousand transitions and gradually become more frequent until the replay buffer is full. Together with my custom exploration policy, this works very well : the agent very quickly starts behaving similar to a “right + random button” policy before learning to actually jump and run. Custom n-step bootstrapping : When I initially implemented n-step bootstrap targets, I initially used n=3 from the Rainbow paper, noting the same instabilities as the paper did for higher n values. I then found the Retrace(\lambda) paper which seems to successfully address this by increasing n until the online network disagrees with the action choice from a stored transition. This makes n larger where the replay buffer data is on-policy, and smaller when it becomes off-policy. Since my GPU is already maxed and the training is already slow (20.8t/s when real-time is 20t/s) I could not afford the additional computations (building a training sample (s(t), a(t), sum(r(t+0…n)), s(t+n)) needs up to n_max transitions to go through the online network). I’m trying to achieve similar sample efficiency gains by using cheaper alternatives as proxies for “how off-policy is a given transition” : I’m using the number of times a transition has been sampled, with n = int(max(n_min, n_max * k**times_sampled)) ; 0<k<1. The currently running experiment uses n_max=14, n_min=1 and k=1/1.3. I’m pretty sure it helps early in the training, and it does not collapse like a constant n=14 does Stream setup : As I said, this is something I do for my own fun, and I really wanted to be able to see the agent learn in real time. The code runs a separate process, to which frames from training episodes are sent in a queue. The process then sends the frames as raw RGB24 to an local UDP socket, to which GStreamer connects and encodes the stream. With a simple MediaMTX configuration, I can manage the Gstreamer process and have the stream available through WebRTC on my LAN. Then I figured someone else might have fun watching this, so I added a line to my MediaMTX config to send the stream to twitch and youtube. The overlay is a headless browser displaying custom HTML/JS (using d3.js for the graphs) piping raw frames to ffmpeg. GStreamer handles compositing the two streams together into the side-by-side view.

publication croisée depuis : lemmy.pierre-couy.fr/post/2152233

I’ve been playing with deep reinforcement learning for a while. I originally started with a simple DQN, added all improvements from the Rainbow paper, and finally changed C51 for a quantile regression (and plan to swap it for an Implicit Quantile Network).

After implementing C51 (which was my first time with distributional RL) I started playing with policies that take advantage of the learned distributions : By independently taking N samples from each action-value distribution, scoring actions by averaging the samples, and picking the greedy action with respect to these scores, I was able to make the agent learn faster than similar agents using only NoisyNets or an epsilon-greedy policy (I’m still using NoisyNet, this is done on top of it). In the limiting cases, N=1 is just Thompson Sampling and N=+Infinity is just a plain greedy policy.

Finding an optimal value for N proved to be a challenge, so I decided to pick a random value for it at the start of each episode (N = 2**rng.uniform(8,12) for a QR-DQN with 32 quantiles/action works well in my experiments), which led to even better results.

I later found out about DLTV which made the agent discover new behaviors, but performed worse than previous experiments overall. Inspired by it, I tried something I did not find in previous works and got the best results out of all my previous experiments :

At each time step, compute an exploration_score as the ratio of “intra-action variance” over “inter-action variance” (rendered latex equation). I then take N/exploration_score samples from each distribution, and pick an action as described above. (more details at the end of this post)

For anyone reading this, I have a few questions :

  1. Are you aware of any previous work I missed that tries similar exploration policies with distributional RL (interpolating between Thompson sampling and the greedy policy)
  2. Most papers I found about learning from multiple exploration policies seem to be in the context of multi-actor parallelization. Is there any novelty in randomizing the policy parameters at the start of each episode, especially in the single-actor case ?
  3. Is any part of what I’m doing worth the time it would take to quantitatively evaluate it ? I’ve been doing it mainly for learning and fun and have only qualitatively evaluated it so far. However, if there’s a chance I can contribute to the field, I’ll gladly make some time to compare it to published papers on ALE.

A few more details

I actually track a moving average and standard deviation of the exploration score, which lets me shift/rescale its values to a target average and standard deviation, and divide N by the shifted/rescaled value. I initially started with a target average of 1 and standard deviation of 1 as well (which gave good results), then tried randomizing these parameters at the start of each episode as well. This led to a lot more diversity in the policies and even better results.

Since this worked so well, I additionally randomized the noise strength in the NoisyNet layers.

Overall, this made the agent a lot more robust to deviating from what it considers to be the optimal trajectory, and allowed it to learn complex behaviors previous iterations were never able to learn (e.g. taking a few steps back to gain momentum, waiting for good cycles, or dodging hammer bros)


Watch it learn

For anyone interested, I made a live stream of the training in progress with graphs and some more details on the experiments I’m running. The current training run was started ~2.5 days ago. The agent has finished and unlocked levels up to 5-1, and is currently learning 5-2.


A lot more details

Long text hidden, click to expand

Available actions : The agent does not have access to the up and down buttons, the available actions only use left, right, A and B. Adding the down button would double the total number of actions (because down can be pressed on top of all available actions). Reward function : It mainly consists of reward(t) = max(0, x(t) - previous_best_x) + a larger reward for beating a stage. I had to tweak the scaling of both components. I initially had penalties for time and death, but one made the agent suicidal in front of hard-to-overcome obstacles, while the other made it fear them too much and hug the left side of the screen. Removing both proved to increase the performance. One trick that seems to help with most ‘*-3’ levels (which have a lot of void to fall into) was to hold the reward while the vertical velocity of Mario is negative (meaning it is falling). Without this trick, the agent would sometimes get stuck learning to jump the farthest it can into the void. Stage scheduling : Each episode is one attempt on one level. At the start of each episode, a stage is randomly picked with probability proportional to 1/(number of times the stage was beaten) among the unlocked stages. Each stage is unlocked after the previous one has been beaten 30 times, with only 1-1 unlocked at the start of the training. Available stages : The first iterations of the agent were unable to learn maze castles (4-3, 7-3 and 8-4), so I removed them all. The reward function will give rewards for the first path the agent tries, then the agent will be teleported back by the game and no reward is received until it finds the right path and gets past the point where the game teleported it back. I plan to test newer (better) versions of the agent on these stages only and see if mazes can be re-added to the pool. I’ve also removed underwater stages (2-2 and 7-2). The agent can learn them fine, but the game dynamics are really different from all other stages and they’re really boring to watch. Since I already removed a bunch of stages, I figured I could remove these as well but I may re-add them with mazes because beating every level is cooler than beating a cherry-picked selection. Since 8-4 is the only stage that requires going down a pipe, I considered it was not worth it to add the down action and will likely never re-add it to the pool, which would unfortunately be really anti-climactic… Replay buffer warm-up : After initially using the standard approach of filling the buffer with transitions sampled from a random policy before training the neural net, I came-up with a “soft warm-up” scheme in which the first gradient updates happen after only 2000 transitions, but initially happen every few thousand transitions and gradually become more frequent until the replay buffer is full. Together with my custom exploration policy, this works very well : the agent very quickly starts behaving similar to a “right + random button” policy before learning to actually jump and run. Custom n-step bootstrapping : When I initially implemented n-step bootstrap targets, I initially used n=3 from the Rainbow paper, noting the same instabilities as the paper did for higher n values. I then found the Retrace(\lambda) paper which seems to successfully address this by increasing n until the online network disagrees with the action choice from a stored transition. This makes n larger where the replay buffer data is on-policy, and smaller when it becomes off-policy. Since my GPU is already maxed and the training is already slow (20.8t/s when real-time is 20t/s) I could not afford the additional computations (building a training sample (s(t), a(t), sum(r(t+0…n)), s(t+n)) needs up to n_max transitions to go through the online network). I’m trying to achieve similar sample efficiency gains by using cheaper alternatives as proxies for “how off-policy is a given transition” : I’m using the number of times a transition has been sampled, with n = int(max(n_min, n_max * k**times_sampled)) ; 0<k<1. The currently running experiment uses n_max=14, n_min=1 and k=1/1.3. I’m pretty sure it helps early in the training, and it does not collapse like a constant n=14 does Stream setup : As I said, this is something I do for my own fun, and I really wanted to be able to see the agent learn in real time. The code runs a separate process, to which frames from training episodes are sent in a queue. The process then sends the frames as raw RGB24 to an local UDP socket, to which GStreamer connects and encodes the stream. With a simple MediaMTX configuration, I can manage the Gstreamer process and have the stream available through WebRTC on my LAN. Then I figured someone else might have fun watching this, so I added a line to my MediaMTX config to send the stream to twitch and youtube. The overlay is a headless browser displaying custom HTML/JS (using d3.js for the graphs) piping raw frames to ffmpeg. GStreamer handles compositing the two streams together into the side-by-side view.

0

The World's Future: How Artificial Intelligence and Humans Collaboration in Shaping Tomorrow

preview

# The World’s Future: How Artificial Intelligence and Humans Collaboration in Shaping Tomorrow

The future has artificial intelligence and humans working hand-in-hand to create it. Combining the two sets us on a path for amazing new breakthroughs. When we step into the possibilities of collaboration between artificial intelligence and humans, it is obvious that this powerful synergy will change our societal norms and potentialities. The idea of “robot dreams” is no longer the exclusive domain of fiction. And today, it stands as a symbol for the promise of AI and human creativity.

The infusion of AI in segments is transforming both frontiers pushing boundaries and innovation dawn new possibilities. The impact by AI on society is significant and diverse. No matter what but be is it healthcare; finance or education and entertainment, AI has streamlined the tasks leading to higher efficiency, accuracy, and personalization.

Such improvements are all part of a wider movement towards the integration between Artificial Intelligence and Humans, living in harmony to solve problems that cannot be achieved by neither side. Organizations like the American Society for AI are emerging to lead efforts in this growing field and encourage ethical standards in it. Their efforts make certain the integration of artificial intelligence and humans benefits society optimally, bringing about a future where technology works for human interest.

Going forward, the partnership between artificial intelligence and humans will undoubtedly make for a more inventive, productive and just world. Together, this untraditional pair are removing the barriers to achieving a future where our robot dreams can become real working in harmony with and for us shrugging of robot security concerns soon. for more read visit : https://freeallai.com/ai-news/artificial-intelligence-and-humans/

# The World’s Future: How Artificial Intelligence and Humans Collaboration in Shaping Tomorrow

The future has artificial intelligence and humans working hand-in-hand to create it. Combining the two sets us on a path for amazing new breakthroughs. When we step into the possibilities of collaboration between artificial intelligence and humans, it is obvious that this powerful synergy will change our societal norms and potentialities. The idea of “robot dreams” is no longer the exclusive domain of fiction. And today, it stands as a symbol for the promise of AI and human creativity.

The infusion of AI in segments is transforming both frontiers pushing boundaries and innovation dawn new possibilities. The impact by AI on society is significant and diverse. No matter what but be is it healthcare; finance or education and entertainment, AI has streamlined the tasks leading to higher efficiency, accuracy, and personalization.

Such improvements are all part of a wider movement towards the integration between Artificial Intelligence and Humans, living in harmony to solve problems that cannot be achieved by neither side. Organizations like the American Society for AI are emerging to lead efforts in this growing field and encourage ethical standards in it. Their efforts make certain the integration of artificial intelligence and humans benefits society optimally, bringing about a future where technology works for human interest.

Going forward, the partnership between artificial intelligence and humans will undoubtedly make for a more inventive, productive and just world. Together, this untraditional pair are removing the barriers to achieving a future where our robot dreams can become real working in harmony with and for us shrugging of robot security concerns soon. for more read visit : https://freeallai.com/ai-news/artificial-intelligence-and-humans/

0

Exclusive: US tackles loopholes in curbs on AI chip exports to China

preview
  • The U.S. is updating rules to prevent American chipmakers from selling AI chips to China that skirt existing restrictions.
  • These updated rules will expand the scope of restricted chips and demand reporting for certain shipments.
  • The move aims to prevent U.S. technology from strengthening China’s military but risks complicating U.S.-China diplomatic efforts.
  • Consumer chips are exempt, but the new rules aim to close loopholes and evolve with technology advancements.
  • The U.S. is updating rules to prevent American chipmakers from selling AI chips to China that skirt existing restrictions.
  • These updated rules will expand the scope of restricted chips and demand reporting for certain shipments.
  • The move aims to prevent U.S. technology from strengthening China’s military but risks complicating U.S.-China diplomatic efforts.
  • Consumer chips are exempt, but the new rules aim to close loopholes and evolve with technology advancements.
0

Google Bard readies ‘Memory’ to adapt to important details about you

9to5Google Google Bard readies ‘Memory’ to adapt to important details about you 9to5google.com
preview

cross-posted from !google@lemdro.id

  • Google’s Bard is developing a “Memory” feature that allows the chatbot to remember and adapt to user-specific details and preferences.
  • Currently, each conversation with Bard starts afresh, requiring users to restate particulars like dietary restrictions.
  • The Memory feature will improve personalized interactions, such as recommending recipes based on a user’s noted vegetarianism or travel suggestions considering the number of kids a user has.
  • Users will be able to manage their stored preferences via a Memory page and can delete incorrect or unwanted memories.
  • A toggle will allow users to easily disable the Memory function for privacy concerns or to have non-personalized conversations.

cross-posted from !google@lemdro.id

  • Google’s Bard is developing a “Memory” feature that allows the chatbot to remember and adapt to user-specific details and preferences.
  • Currently, each conversation with Bard starts afresh, requiring users to restate particulars like dietary restrictions.
  • The Memory feature will improve personalized interactions, such as recommending recipes based on a user’s noted vegetarianism or travel suggestions considering the number of kids a user has.
  • Users will be able to manage their stored preferences via a Memory page and can delete incorrect or unwanted memories.
  • A toggle will allow users to easily disable the Memory function for privacy concerns or to have non-personalized conversations.
0

What to Know About Worldcoin and the Controversy Around It

TIME What to Know About Worldcoin and the Controversy Around It time.com
preview

By Andrew R. Chow

Over the past few months, shiny metallic orbs have materialized cities around the world, from New York to Berlin to Tokyo. Its creators hail the orbs as revolutionary devices ushering in a new era of global humanity and financial stability. Its detractors slam them as invasive, dystopian and exploitative.

Welcome to the rollout of Worldcoin, an AI-meets-crypto project from OpenAI founder Sam Altman that has stirred endless controversy. The startup uses orbs to scan people’s eyes in exchange for a digital ID and possibly some cryptocurrency, depending on what country they live in.

Altman and his co-founder Alex Blania hope that Worldcoin will provide a new solution to online identity in a digital landscape rife with scams, bots and even AI imposters. But privacy experts are concerned about the Worldcoin’s collection of biometric data and how, exactly, the project will keep and protect that data going forward. On Aug. 2, Kenya became the first country to suspend Worldcoin’s activities and its government launched a multi-agency investigation into the project’s practices.

Here’s what to know about the new technology. What is the aim of Worldcoin?

Navigating the online world in 2023 can often feel like an endless obstacle course of lurking dangers. Many websites require a login and password. Scammers have all sorts of strategies to get you to click on links and send them money. Bots run rampant on social media platforms.

And with the rise of AI, our collective ability to discern who is a human online and who isn’t is about to become much worse. Earlier this year, OpenAI’s GPT-4 was even able to convince a human to solve a CAPTCHA—a technology designed to differentiate humans from bots—on its behalf.

As a founder and the CEO of OpenAI, Sam Altman bears responsibility for problems like this. Worldcoin, then, serves as his potential solution: a way to definitively distinguish between humans and AIs. If all humans online could prove that they were, in fact, humans, then scams and imposters would dramatically decrease, and the digital landscapes would become more accurate representations of us as a society.

So in order to prove that humans are humans, Worldcoin scans irises, which are unique to their owners. This technique is not unlike the biometric scans conducted by CLEAR or Apple’s Face ID.

Once Worldcoin has received a unique iris scan, the project issues a digital identity called a World ID. The ID is not a user’s biometric data itself, but an identifier created using a cryptography method called zero-knowledge proofs.

If World IDs catch on, then holders could theoretically use them to sign on to all websites, just like Google offers single sign-on services. The difference, Altman argues, is that a Worldcoin login will be more secure and unlinked to other information, including a user’s email, name or photograph. If the zero-knowledge proof aspect of the tech works properly, Worldcoin would allow ID holders to log into a website without that action being traceable by other people or any government.

In June, Okta became the first major company to allow users to sign in with Worldcoin. Why are people signing up?

Worldcoin officially launched in July, with the project embarking on a multi-city sign-up tour. Altman posted a video of long lines outside Orb centers, and said that the project was scanning in a new user every eight seconds.

Worldcoin claims that more than two million people have registered for World IDs. Some people have signed up out of curiosity or affinity for the technology. But many more, it seems, signed up for a simpler reason: money. In order to galvanize interest in the project, Worldcoin created its own crypto token, called WLD, and has offered 25 units (currently worth about $60) to anyone who scanned their eyes into an orb. That offer does not extend to the U.S., however, due to the country’s strict regulatory environment related to crypto products.

Worldcoin’s cryptocurrency element serves as a marketing ploy, a global financial infrastructure, and a way to entice continued venture capital funding. (Tools for Humanity, the company behind Worldcoin, just raised $115 million in a funding round and was valued at $3 billion last year. Around 14% of all WLD tokens have been earmarked for investors.) Worldcoin, which Altman says he conceived of in 2019, has for years been entwined with the crypto ecosystem, for better or worse. It was publicly announced in the summer of 2021, when crypto was near its height, and the project’s original backers include Sam Bankman-Fried, the former CEO of FTX, accused of perpetrating an $8 billion fraud, and the collapsed crypto hedge fund Three Arrows Capital.

Sam Altman says he hopes that Worldcoin’s financialized element could lead to universal basic income (UBI) for its users. Altman has long been interested in UBI. When he was the president of startup accelerator Y Combinator, he led a pilot trial that aimed to give $1,500 a month to Oakland families. But the project was significantly delayed and reduced in scale. Elizabeth Rhodes, project director of YC Research—the incubator’s nonprofit arm, which has since spun off—told Wired in 2018 that “it’s harder to give away money than you might think.”

Worldcoin also isn’t the first project to entwine digital identity with universal basic income. Last year, a project called Proof of Humanity attempted something similar—and for a while, was handing out $50 to $100 monthly in crypto to everyone who signed up. But the token’s value cratered along with the larger market.

Santiago Siri, a board member of Proof of Humanity, says that Worldcoin’s cryptographic mechanisms designed to protect identity represent a step forward from his project, which required new users to post a public video to verify their identity. Siri hopes that Altman will leverage Worldcoin and his growing revenue from OpenAI to redistribute wealth. “AI can certainly hold true the promise of delivering a global universal basic income, because it can generate that kind of sustainable wealth throughout time. And because it will displace jobs, they have the ethical obligation to actually support UBI initiatives,” he says.

How Altman plans to pay for Worldcoin’s UBI, however, remains murky. “The hope is that as people want to buy this token, because they believe this is the future, there will be inflows into this economy. New token-buyers is how it gets paid for, effectively,” Altman told Coindesk recently.

While Altman hopes that the price of WLD will go up, its value so far has been somewhat volatile. WLD debuted at $7.50 but has since dropped threefold, and has hovered between $2 and $3 over the last week. Why is Worldcoin facing criticism?

Many critics have called Worldcoin’s business—of scanning eyeballs in exchange for crypto—dystopian and some have compared it to bribery. Many people are understandably hesitant to surrender their biometric data to a private for-profit startup with uncertain aims; even Altman himself acknowledged a “clear ick factor.”

Worldcoin says the biometric information on the orbs is deleted after being processed and converted into cryptographic code. But a history of Silicon Valley companies mishandling data has left a sour taste in peoples’ mouths, and some fear that the iris scans could be used for surveillance or be sold to third parties. “Don’t use biometrics for anything,” Edward Snowden wrote on Twitter in response to Altman’s post about Worldcoin in 2021. “The human body is not a ticket-punch.”

Meanwhile, Worldcoin’s rollout has been plagued by troubling reports from around the world. An extensive 2022 article from the MIT Technology Review found evidence that the project used deceptive practices to sign people up in countries like Indonesia, Kenya and Chile. Spanish-language speakers, for instance, were given terms of service notices in English, and people in Sudan were enticed with AirPod giveaways without being told what exactly they were scanning their eyeballs for.

A representative for the Worldcoin Foundation responded to the Technology Review article in a statement to TIME, writing: “The article is not representative of the project’s operations, included inaccurate information and offered a narrow, personal perspective of the incredibly early stages of the project. It is also not an accurate representation of the project’s global operations today.”

This year, hackers stole the login credentials of Worldcoin operators who are tasked with signing up new users, allowing the hackers to view internal information. An article by the crypto publication BlockBeats alleged that people in Cambodia and Kenya were selling their iris data for as little as $30 a pop to speculators on the black market, who hoped that the WLD they collected from the scans would increase in price.

“I think that there are important questions to be asked about whether or not this opens the path towards artificial intelligence colonialism,” Proof of Humanity’s Siri, who is Argentinian, says. “We have seen the orb being deployed in third-world developing countries whose rules about identity and privacy might not be as strong as they are in the European Union or in the United States.”

Regulators around the world have been watching Worldcoin’s rise closely. The French data protection watchdog announced a probe into the project over its “questionable” data collection and preservation. A U.K. regulator issued a similar warning. And the Kenyan government demanded Worldcoin cease its data collection activities there, writing the project posed “legitimate regulatory concerns that require urgent action.”

A Worldcoin Foundation representative responded to the news in Kenya in a statement to TIME, writing: “The demand for Worldcoin’s proof of personhood verification services in Kenya has been overwhelming, resulting in tens of thousands of individuals waiting in lines over a three-day period to secure a World ID…Worldcoin remains committed to providing an inclusive, privacy-preserving, decentralized on-ramp to the global digital economy and looks forward to resuming its services in Kenya while working closely with local regulators and other stakeholders.”

When Worldcoin co-founder Alex Blania was asked by Bloomberg News about login theft and black market sales, he dismissed their impact. “Of course there will be fraud,” he said. “It will not be a perfect system, especially early on.”

By Andrew R. Chow

Over the past few months, shiny metallic orbs have materialized cities around the world, from New York to Berlin to Tokyo. Its creators hail the orbs as revolutionary devices ushering in a new era of global humanity and financial stability. Its detractors slam them as invasive, dystopian and exploitative.

Welcome to the rollout of Worldcoin, an AI-meets-crypto project from OpenAI founder Sam Altman that has stirred endless controversy. The startup uses orbs to scan people’s eyes in exchange for a digital ID and possibly some cryptocurrency, depending on what country they live in.

Altman and his co-founder Alex Blania hope that Worldcoin will provide a new solution to online identity in a digital landscape rife with scams, bots and even AI imposters. But privacy experts are concerned about the Worldcoin’s collection of biometric data and how, exactly, the project will keep and protect that data going forward. On Aug. 2, Kenya became the first country to suspend Worldcoin’s activities and its government launched a multi-agency investigation into the project’s practices.

Here’s what to know about the new technology. What is the aim of Worldcoin?

Navigating the online world in 2023 can often feel like an endless obstacle course of lurking dangers. Many websites require a login and password. Scammers have all sorts of strategies to get you to click on links and send them money. Bots run rampant on social media platforms.

And with the rise of AI, our collective ability to discern who is a human online and who isn’t is about to become much worse. Earlier this year, OpenAI’s GPT-4 was even able to convince a human to solve a CAPTCHA—a technology designed to differentiate humans from bots—on its behalf.

As a founder and the CEO of OpenAI, Sam Altman bears responsibility for problems like this. Worldcoin, then, serves as his potential solution: a way to definitively distinguish between humans and AIs. If all humans online could prove that they were, in fact, humans, then scams and imposters would dramatically decrease, and the digital landscapes would become more accurate representations of us as a society.

So in order to prove that humans are humans, Worldcoin scans irises, which are unique to their owners. This technique is not unlike the biometric scans conducted by CLEAR or Apple’s Face ID.

Once Worldcoin has received a unique iris scan, the project issues a digital identity called a World ID. The ID is not a user’s biometric data itself, but an identifier created using a cryptography method called zero-knowledge proofs.

If World IDs catch on, then holders could theoretically use them to sign on to all websites, just like Google offers single sign-on services. The difference, Altman argues, is that a Worldcoin login will be more secure and unlinked to other information, including a user’s email, name or photograph. If the zero-knowledge proof aspect of the tech works properly, Worldcoin would allow ID holders to log into a website without that action being traceable by other people or any government.

In June, Okta became the first major company to allow users to sign in with Worldcoin. Why are people signing up?

Worldcoin officially launched in July, with the project embarking on a multi-city sign-up tour. Altman posted a video of long lines outside Orb centers, and said that the project was scanning in a new user every eight seconds.

Worldcoin claims that more than two million people have registered for World IDs. Some people have signed up out of curiosity or affinity for the technology. But many more, it seems, signed up for a simpler reason: money. In order to galvanize interest in the project, Worldcoin created its own crypto token, called WLD, and has offered 25 units (currently worth about $60) to anyone who scanned their eyes into an orb. That offer does not extend to the U.S., however, due to the country’s strict regulatory environment related to crypto products.

Worldcoin’s cryptocurrency element serves as a marketing ploy, a global financial infrastructure, and a way to entice continued venture capital funding. (Tools for Humanity, the company behind Worldcoin, just raised $115 million in a funding round and was valued at $3 billion last year. Around 14% of all WLD tokens have been earmarked for investors.) Worldcoin, which Altman says he conceived of in 2019, has for years been entwined with the crypto ecosystem, for better or worse. It was publicly announced in the summer of 2021, when crypto was near its height, and the project’s original backers include Sam Bankman-Fried, the former CEO of FTX, accused of perpetrating an $8 billion fraud, and the collapsed crypto hedge fund Three Arrows Capital.

Sam Altman says he hopes that Worldcoin’s financialized element could lead to universal basic income (UBI) for its users. Altman has long been interested in UBI. When he was the president of startup accelerator Y Combinator, he led a pilot trial that aimed to give $1,500 a month to Oakland families. But the project was significantly delayed and reduced in scale. Elizabeth Rhodes, project director of YC Research—the incubator’s nonprofit arm, which has since spun off—told Wired in 2018 that “it’s harder to give away money than you might think.”

Worldcoin also isn’t the first project to entwine digital identity with universal basic income. Last year, a project called Proof of Humanity attempted something similar—and for a while, was handing out $50 to $100 monthly in crypto to everyone who signed up. But the token’s value cratered along with the larger market.

Santiago Siri, a board member of Proof of Humanity, says that Worldcoin’s cryptographic mechanisms designed to protect identity represent a step forward from his project, which required new users to post a public video to verify their identity. Siri hopes that Altman will leverage Worldcoin and his growing revenue from OpenAI to redistribute wealth. “AI can certainly hold true the promise of delivering a global universal basic income, because it can generate that kind of sustainable wealth throughout time. And because it will displace jobs, they have the ethical obligation to actually support UBI initiatives,” he says.

How Altman plans to pay for Worldcoin’s UBI, however, remains murky. “The hope is that as people want to buy this token, because they believe this is the future, there will be inflows into this economy. New token-buyers is how it gets paid for, effectively,” Altman told Coindesk recently.

While Altman hopes that the price of WLD will go up, its value so far has been somewhat volatile. WLD debuted at $7.50 but has since dropped threefold, and has hovered between $2 and $3 over the last week. Why is Worldcoin facing criticism?

Many critics have called Worldcoin’s business—of scanning eyeballs in exchange for crypto—dystopian and some have compared it to bribery. Many people are understandably hesitant to surrender their biometric data to a private for-profit startup with uncertain aims; even Altman himself acknowledged a “clear ick factor.”

Worldcoin says the biometric information on the orbs is deleted after being processed and converted into cryptographic code. But a history of Silicon Valley companies mishandling data has left a sour taste in peoples’ mouths, and some fear that the iris scans could be used for surveillance or be sold to third parties. “Don’t use biometrics for anything,” Edward Snowden wrote on Twitter in response to Altman’s post about Worldcoin in 2021. “The human body is not a ticket-punch.”

Meanwhile, Worldcoin’s rollout has been plagued by troubling reports from around the world. An extensive 2022 article from the MIT Technology Review found evidence that the project used deceptive practices to sign people up in countries like Indonesia, Kenya and Chile. Spanish-language speakers, for instance, were given terms of service notices in English, and people in Sudan were enticed with AirPod giveaways without being told what exactly they were scanning their eyeballs for.

A representative for the Worldcoin Foundation responded to the Technology Review article in a statement to TIME, writing: “The article is not representative of the project’s operations, included inaccurate information and offered a narrow, personal perspective of the incredibly early stages of the project. It is also not an accurate representation of the project’s global operations today.”

This year, hackers stole the login credentials of Worldcoin operators who are tasked with signing up new users, allowing the hackers to view internal information. An article by the crypto publication BlockBeats alleged that people in Cambodia and Kenya were selling their iris data for as little as $30 a pop to speculators on the black market, who hoped that the WLD they collected from the scans would increase in price.

“I think that there are important questions to be asked about whether or not this opens the path towards artificial intelligence colonialism,” Proof of Humanity’s Siri, who is Argentinian, says. “We have seen the orb being deployed in third-world developing countries whose rules about identity and privacy might not be as strong as they are in the European Union or in the United States.”

Regulators around the world have been watching Worldcoin’s rise closely. The French data protection watchdog announced a probe into the project over its “questionable” data collection and preservation. A U.K. regulator issued a similar warning. And the Kenyan government demanded Worldcoin cease its data collection activities there, writing the project posed “legitimate regulatory concerns that require urgent action.”

A Worldcoin Foundation representative responded to the news in Kenya in a statement to TIME, writing: “The demand for Worldcoin’s proof of personhood verification services in Kenya has been overwhelming, resulting in tens of thousands of individuals waiting in lines over a three-day period to secure a World ID…Worldcoin remains committed to providing an inclusive, privacy-preserving, decentralized on-ramp to the global digital economy and looks forward to resuming its services in Kenya while working closely with local regulators and other stakeholders.”

When Worldcoin co-founder Alex Blania was asked by Bloomberg News about login theft and black market sales, he dismissed their impact. “Of course there will be fraud,” he said. “It will not be a perfect system, especially early on.”

0

Early adopters in Mexico lend their eyes to global biometric project

preview

Eager early adopters recently descended upon a Mexico City cafe where their eyes were scanned by a futuristic sphere, part of an ambitious project that ultimately seeks to create a unique digital identification for everyone on the planet.

Mexico is one of nearly three dozen countries where participants are allowing the sphere, outfitted with cameras and dubbed an orb, to scan their iris. The project’s goal is to distinguish people from bots online, while doling out a cryptocurrency bonus as a incentive to participate.

The so-called Worldcoin project is a biometric verification tool led by Sam Altman, the chief executive of Open AI, and the crypto company he co-founded, Tools for Humanity.

Not directly AI related but the Sam Altman tie-in is interesting. This doesn’t sound creepy at all… thoughts?

Eager early adopters recently descended upon a Mexico City cafe where their eyes were scanned by a futuristic sphere, part of an ambitious project that ultimately seeks to create a unique digital identification for everyone on the planet.

Mexico is one of nearly three dozen countries where participants are allowing the sphere, outfitted with cameras and dubbed an orb, to scan their iris. The project’s goal is to distinguish people from bots online, while doling out a cryptocurrency bonus as a incentive to participate.

The so-called Worldcoin project is a biometric verification tool led by Sam Altman, the chief executive of Open AI, and the crypto company he co-founded, Tools for Humanity.

Not directly AI related but the Sam Altman tie-in is interesting. This doesn’t sound creepy at all… thoughts?

0

Google, Universal Music in talks for deal on AI 'deepfakes' - FT

preview

Aug 8 (Reuters) - Alphabet’s (GOOGL.O) Google and Universal Music (UMG.AS) are in talks to license artists’ voices and melodies for artificial intelligence-generated songs, Financial Times reported on Tuesday, citing four people familiar with the matter.

The music industry is grappling with “deepfake” songs, made using generative AI, that mimic artists’ voices, often without their consent.

The goal behind the talks is to develop a tool for fans to create tracks legitimately and pay the owners of the copyrights for them, the report said, adding the artists would have a choice to opt in the process.

Discussions between Google and Universal Music are at an early stage and no product launch is imminent, while Warner Music (WMG.O) is also in talks with Google about a product, the report added.

The companies did not immediately respond to Reuters’ requests for comment.

Aug 8 (Reuters) - Alphabet’s (GOOGL.O) Google and Universal Music (UMG.AS) are in talks to license artists’ voices and melodies for artificial intelligence-generated songs, Financial Times reported on Tuesday, citing four people familiar with the matter.

The music industry is grappling with “deepfake” songs, made using generative AI, that mimic artists’ voices, often without their consent.

The goal behind the talks is to develop a tool for fans to create tracks legitimately and pay the owners of the copyrights for them, the report said, adding the artists would have a choice to opt in the process.

Discussions between Google and Universal Music are at an early stage and no product launch is imminent, while Warner Music (WMG.O) is also in talks with Google about a product, the report added.

The companies did not immediately respond to Reuters’ requests for comment.

0

Microsoft’s AI Red Team Has Already Made the Case for Itself

WIRED Microsoft’s AI Red Team Has Already Made the Case for Itself www.wired.com
preview

Since 2018, a dedicated team within Microsoft has attacked machine learning systems to make them safer. But with the public release of new generative AI tools, the field is already evolving.

Since 2018, a dedicated team within Microsoft has attacked machine learning systems to make them safer. But with the public release of new generative AI tools, the field is already evolving.

0

Dungeons &amp; Dragons tells illustrators to stop using AI to generate artwork for fantasy franchise

preview

The Dungeons & Dragons role-playing game franchise says it won’t allow artists to use artificial intelligence technology to draw its cast of sorcerers, druids and other characters and scenery.

The Dungeons & Dragons role-playing game franchise says it won’t allow artists to use artificial intelligence technology to draw its cast of sorcerers, druids and other characters and scenery.