A Look at Different Countries' Approaches to AI Regulations - H+ Weekly - Issue #413
This week - Geoffrey Hinton quits Google to focus on AI safety; mind-reading LLMs; real-world sci-fi war room; robots learn to play soccer and go fast; self-repairing vegan leather; and more!
I have a short announcement before we jump into this week’s issue. I’m planning to launch a new project involving racing robots to promote science and engineering education and I’m building a team to launch this project with. If that sounds interesting to you, please send me a message.
Generative AIs such as GPT-4 or Midjourney have been one of the main topics of discussion this year. Some people raised concerns seeing how powerful they are and called to introduce legal frameworks to regulate their usage.
The European Union has a very strict approach to privacy and it was the first place where generative AIs got in trouble. Italy was the first country to temporarily ban ChatGPT over privacy concerns at the end of March. Germany was also considering a similar ban while France and Spain were investigating OpenAI over a suspected breach of privacy rules. The ChatGPT’s ban in Italy has been lifted recently after OpenAI improved “transparency and rights for European users."
EU is paving the way for the world's first comprehensive laws governing AI. On April 27th, the members of the European Parliament agreed to move the AI Act to the next stage where the EU lawmakers and member states will sort out the final details. The proposed rules will classify AI tools into four risk categories - unacceptable, high, limited, and minimal - each of which triggers different regulatory requirements. Areas of concern could include biometric surveillance, deepfakes, spreading misinformation or discriminatory language.
The EU will also reinforce their stance on AI transparency by requiring AI companies to disclose any copyrighted material used in developing and training their products.
The proposed EU AI Act can be viewed here. The draft bill is not final and it could take years to come into force.
China is another highly regulated market when it comes to software, AI and internet usage. ChatGPT is banned in China in favour of local products and services.
The proposed rules will require AI companies to make sure their products “reflect the core values of socialism” and the generated content is “true and accurate”. The responsibility for the generated content will be placed on the service provider who will be also responsible for “the legitimacy of the source of pre-training data and optimization training data for generative artificial intelligence products”. Under new rules, any service using generative AI will have to “undergo security assessments and receive government approval before being offered to users”. Providers who fail to comply with new rules will be fined, have their services suspended, or even face criminal investigations.
Not everyone is rushing to regulate AI and AI-generated content.
In March, the UK Department for Science, Innovation and Technology (DSIT) released a white paper outlining a “A pro-innovation approach to AI regulation”. The UK government has chosen to promote innovation over regulations and won’t be introducing any new AI-specific rules anytime soon in favour of context-specific guidance and flexible principles.
For a long time, the US took a hands-off approach to regulating AI. As New York Times stated in this article, there were some initiatives such as the White House’s Blueprint for an AI Bill of Rights or AI Risk Management Framework but none of them turned into law. The federal agencies take some actions around AI but they are enforcing existing laws.
Recently, the White House took the initiative by inviting CEOs of Microsoft, Google and OpenAI to discuss key artificial intelligence issues. The invitation seen by Reuters to the CEOs noted President Joe Biden's "expectation that companies like yours must make sure their products are safe before making them available to the public."
Yesterday, May 4th, the White House released a document announcing “new actions to promote responsible AI innovation that protects Americans’ rights and safety”. The announcements include new “investments to power responsible American AI research and development, “an independent commitment from leading AI developers, including Anthropic, Google, Hugging Face, Microsoft, NVIDIA, OpenAI, and Stability AI, to participate in a public evaluation of AI systems” and “policies to ensure the U.S. government is leading by example on mitigating AI risks and harnessing AI opportunities”.
From H+ Weekly
🦾 More than a human
Brain scans can translate a person’s thoughts into words
Researchers have trained a large language model to turn thoughts into words. What they did is they asked people to listen to podcasts while they scanned their brain activity with fMRI. The fMRI data combined with what the participants were listening to was used to train an LLM-based AI to translate thoughts into words. According to the researchers, the AI was able to recover the gist of what users were hearing just from their brain activity, often identifying exact words and phrases.
Inside the Secretive Life-Extension Clinic
Wired shares a story about BioViva - a company working on extending the human lifespan, whose founder and CEO, Liz Parrish, gained notoriety for claiming to be the first ever person to get an age-reversing gene therapy. A claim that hasn’t been independently verified yet. But that does not stop Parrish or BioViva from applying the “move fast and break things” mantra to medicine and offering gene therapies to reverse ageing outside the established medical industry.
This AI-powered exoskeleton does the heavy lifting so you don’t have to
Exoskeletons are here and are being used in real-life scenarios. One of the companies in this space is German Bionic which is featured in this article. Their products are being used by companies such as UK tech retailer Currys and German parcel delivery service DPD to help their workers doing physically demanding jobs avoid injuries.
▶️ How a brain prosthesis could help the blind (10:27)
In this TEDx Talk, Pieter Roelfsema shares his research in creating brain implants to restore vision and demonstrates how it is possible to directly convey information to the parts of the brain of people whose eyes stopped working. The idea is to take the information from a camera that the blind person wears and to directly plug it into the visual cortex. It is not a full vision yet (it’s more like a cloud of dots) but his idea has been proven to work in monkeys.
🧠 Artificial Intelligence
Deep learning pioneer Geoffrey Hinton quits Google
Geoffrey Hilton, the AI pioneer who massively contributed to the deep learning explosion, one of the "Godfathers of AI" and "Godfathers of Deep Learning", is leaving Google after 10 years to focus on AI safety. In this conversation with MIT Technology Review, Hilton says he is scared of what he has helped to create and how he wants to make sure AI won’t destroy humanity. One of Hinton’s priorities is to try to work with leaders in the technology industry to see if they can come together and agree on what the risks are and what to do about them. Bernie Sanders, Elon Musk and White House have apparently already reached out to him.
Republicans respond to Biden reelection announcement with AI-generated attack ad
With generative AIs and tools to create deepfakes being easy to access and use, the 2024 US Presidential campaign is going to be interesting. We’ve just got a glimpse of propaganda and misinformation heading our way with this video using AI-generated images the Republican National Committee released in response to Joe Biden announcing his reelection campaign.
▶️ Palantir AIP | Defense and Military (8:05)
Palantir presents AIP - their military-focused application of large language models. In this demo, they show how AIP and LLMs can be used in a fictional battle and how they can deploy a drone to collect more information, identify threats, create responses, prepare orders for ground units and help execute these orders. What Palantir presents here is basically the sci-fi war room becoming a reality.
▶️ ChatGPT in Skyrim VR (6:41)
I did not expect to see these three words - ChatGPT, Skyrim and VR - next to each other but here we are. Someone created a mod for Skyrim that combines ChatGPT, xVASynth (text-to-speech), and Whisper (speech-to-text) to bring as natural as possible conversations with NPCs. Some game designers were speculating on how generative AI can be used in gaming and here we have a good example of it.
Little Robots Learn to Drive Fast in the Real World
By combining pre-trained models of robots driving around and reinforced learning, roboticists from UC Berkley were able to teach a tiny rally robot to quickly learn how to go fast. With their method, the robot taught itself to match human performance after just 20 minutes of practice.
Learning Agile Soccer Skills for a Bipedal Robot with Deep Reinforcement Learning
Researchers from DeepMind share how they applied deep reinforcement learning to teach small and affordable OP3 humanoid robots to play soccer. It is fascinating to see how these robots went from nothing to being quite good at the game and how they were able to adapt the lessons learned in a simulated environment to the real world in a 1v1 game.
▶️ Robert Playter: Boston Dynamics CEO on Humanoid and Legged Robotics (2:27:57)
In this video, Lex Fridman interviews Robert Playter - co-founder and CEO of Boston Dynamics - and asks him to share how it was building one of the most important robotics company in history, from early experiments in the basement of MIT to capturing people’s attention with humanoid robots doing parkour. It is a fascinating conversation in which Playter (who has the ability to explain complex engineering challenges in a simple and engaging way) shares the behind the scene stories from Boston Dynamics and what they learned from building advanced robots and turning research and development projects into viable robotics business.
Gene therapy reverses vision loss in primates — by making their eyes young again
A new gene therapy that reprograms cells to their younger state reversed vision loss in primates, according to a new study — potentially signalling a new era in the treatment of age-related diseases.
Gel cures 100% of mice with deadly brain cancer
A team of researchers has combined an anticancer drug and an antibody in a solution that self-assembles into a gel to fill the tiny grooves left after a brain tumour is surgically removed. The gel can reach areas that surgery might miss and current drugs struggle to reach to kill lingering cancer cells and suppress tumour growth. The gel appears to not only fend off cancer but help rewire the immune system to discourage recurrence with immunological memory, the researchers say.
A vegan leather made of dormant fungi can repair itself
Leather made from fungi is not anything new. But a fungi leather that can regrow itself is something new. Bioengineers have found a way to grow sheets of vegan leather without killing the mycelium, allowing it to repair itself. The technique could potentially go beyond a proof-of-concept and into commercialisation in the next decade. But before that happens, more research needs to be done to improve the leather’s strength and to make sure it won’t dissolve or start growing uncontrollably.
H+ Weekly is a free, weekly newsletter with the latest news and articles about AI, robotics, biotech and technologies that blur the line between humans and machines, delivered to your inbox every Friday.
Subscribe to H+ Weekly to support the newsletter under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).
This issue is also supported by my Patreons: whmr, Florian, dux, Eric and Andrew. Thank you for your support!
Thank you for reading and see you next Friday!
H+ Weekly is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.