

Discover more from Last Week in AI
Last Week in AI #197: AI for Drug Discovery, Deep Reinforcement Learning for Atom Manipulation, SF rolls back plan for killer police robots, and more!
The tech underpinning ChatGPT can also be used for drug discovery, deep RL-controlled nano robot that can build materials atom by atom, SF reverses decision for police use of killer robots
Top News
How AI That Powers Chatbots and Search Queries Could Discover New Drugs
Recent advances in natural language processing have enabled exciting products like ChatGPT and improvements to existing systems like search. What are the less obvious applications of this technology? You might recall from your biology classes that proteins--the important molecules that underly the structure and function of the body and are used in drugs--are themselves made up of amino acids. The biological codes for these proteins are sequences of letters, where the letters correspond to amino acids. NLP can be applied to predicting sequences of letters, which could themselves become the basis of new drug molecules--the time required for the early stages of drug discovery could shrink drastically. While NLP-based drug discovery is in its early stages and has many challenges ahead, there is a great deal of interest, and companies, for now, are using protein-language models to enhance known molecules for applications like improving the efficacy of drug candidates.
Our Take: Our very own Daniel Bashir actually spoke to investor and State of AI Report author Nathan Benaich about these exact sorts of applications not too long ago: he's excited about the applications of diffusion models to biology and chemistry. I (Daniel) too think these second-order applications are going to be where we really start to hit it out of the park and expect that NLP for predicting amino acid sequences will just be the start.
The smallest robotic arm you can imagine is controlled by artificial intelligence
Researchers used deep reinforcement learning (deep RL) to steer single silver atoms into a star-like lattice. This process is similar to moving marbles around a Chinese checkers board. Applications for this technology include building nanodevices (transistors, memory, etc) and testing how they work at the nanoscale, as well as building new materials atom-by-atom. It took the algorithm on the order of one day to learn and then about one hour to build the lattice. This approach could accelerate the parameter selection and trial-and-error usually done by a person in nanotechnique experiments.
Our take: In past years deep RL has gotten a bad rep for being data hungry, unreliable, and not usable in actual applications. However, this take is really becoming worse and worse as labs and companies across the world demonstrate successful deep RL applications. Deep RL shines in settings where the exact systems are hard to model, and nanotech is one such setting, as this work demonstrates. We’ve also seen deep RL successes in controlling fusion reactors, and RL is core to the success of ChatGPT. These all make me optimistic about RL and more impactful applications down the line.
San Francisco reverses plans to allow police robots to kill suspects
San Francisco’s board of supervisors has reversed a policy that would have allowed police to use remote-controlled robots to kill suspects. The policy was met with outcry from citizens, civil rights groups, and advocacy groups, who argued it would lead to an increase in violence and the militarization of law enforcement. The policy was originally approved but has now been sent back to a committee for further discussion. Supervisor Gordon Mar, who had originally voted in favor of the policy, changed his stance, saying he was uncomfortable with the precedent it set. The reversal was celebrated by Supervisor Hillary Ronen, who had previously voted against the policy.
Our take: This is a great example of civic action mitigating potentially harmful uses of technology. SF’s board of supervisors showed courage and conviction in reversing their initial decision, in spite of the pressure from law enforcement to approve the policy. It is a testament to the power of public outcry and demonstrates the importance of holding our elected officials accountable. This reversal sets an important precedent for other cities that may also be considering the deployment of killer robots.
Other News
Research
Robot Learns Human Trick for Not Falling Over - "Humanoid limbs are useful for more than just manipulation"
OpenAI’s ChatGPT shows why implementation is key with generative AI - "It’s probably not a secret to those doing a lot of focused work in the space, but when it comes to generative AI, it’s quickly becoming apparent that how a user interfaces with generative models and systems is at least as important as the underlying training and inference technology."
AI Enables the Largest Brain Tumor Study To-Date, Led by Penn - "Penn led the largest global effort to identify and predict GBM tumor boundaries without compromising patient privacy."
DeepMind & UCL Fine-tune a 70B Parameter LM to Generate Statements Agreeable to Humans with Diverse Opinions - "LLM to generate statements that maximize agreement among a human group with diverse written opinions"
Soft robot detects damage, heals itself - "Cornell University engineers have created a soft robot capable of detecting when and where it was damaged -- and then healing itself on the spot."
AI learns to write computer code in ‘stunning’ advance - "DeepMind’s AlphaCode outperforms many human programmers in tricky software challenges"
DeepMind created an AI tool that can help generate rough film and stage scripts - "Have you ever thought up an idea for a movie or play that you just know will be a smash hit, but haven't gotten around to writing the script? Alphabet's DeepMind has built an AI tool that can help get you started."
NIH funds Illinois Tech's machine learning-based insulin delivery - "Professor of chemical engineering Ali Cinar leads the project aimed at easing the burden on those with type 1 diabetes. NIH funding covers for years for developing the machine learning system. The researchers aim to integrate it into Cinar’s artificial pancreas system to enhance accuracy."
Applications
A new film festival will only show movies made using AI - "Runway ML is inviting creators of short films made with artificial intelligence to compete for a $10K prize."
Apple slices its AI image synthesis times in half with new Stable Diffusion fix - "On Wednesday, Apple released optimizations that allow the Stable Diffusion AI image generator to run on Apple Silicon using Core ML, Apple's proprietary framework for machine learning models."
This AI Chatbot Is Blowing People’s Minds. Here’s What It’s Been Writing. - "From historical arguments to poems on cryptocurrency, users took to Twitter to share their surprise at the detailed answers the so-called ChatGPT provided, after the startup sought user feedback on the AI model Wednesday."
Would you watch these cheesy holiday movies written by AI? - "Holiday movies bring comfort and tug at heartstrings in the cheesiest ways. So in our tech obsessed culture what would happen if AI tried to take over and write a classic holiday movie?"
Predicting European cities climate mitigation performance using machine learning - "Although cities have risen to prominence as climate actors, emissions data scarcity has been the primary challenge to evaluating their performance."
AI spots deepfake videos of Ukrainian president Volodymyr Zelenskyy - "A deepfake detector designed to identify unique facial expressions and hand gestures could spot manipulated videos of world leaders such as Volodymyr Zelenskyy and Vladimir Putin"
Business
Uber, Aurora to expand self-driving truck ops in Texas to meet holiday rush - "The program will be expanded to the recently launched 600-mile commercial lane between Fort Worth and El Paso in Texas to support customers of logistics business Uber Freight as it ships goods this holiday season, Aurora said on Friday."
Apple Scales Back Self-Driving Car and Delays Debut Until 2026 - "Apple Inc. has scaled back ambitious self-driving plans for its future electric vehicle and postponed the car’s target launch date by about a year to 2026, according to people with knowledge of the matter. "
The global citizen services AI market is projected to grow from USD 6.5 Billion in 2022 to USD 41.0 Billion by 2027, at a CAGR of 44.5% - "during the forecast period. Governments across the globe had begun to adopt cutting-edge technologies, such as artificial intelligence (AI) and machine learning (ML), to hasten the delivery of a number of citizen services."
Publisher releases children’s book with AI-generated illustrations - "Poughquag-based Stone Hollow Press is releasing a new children's book featuring illustrations created with artificial intelligence (AI)."
Google shuts down Duplex on the Web, its attempt to bring AI smarts to retail sites and more - "Google is shutting down Duplex on the Web, its AI-powered set of services that navigated sites to simplify the process of ordering food, purchasing movie tickets, changing compromised passwords and more."
Locus Robotics raises $117 million in new funding, valuing startup at $2 billion - "Locus Robotics, a maker of autonomous mobile robots for fulfillment and distribution warehouses, has raised $117 million in Series F funding, led by Goldman Sachs Asset Management and G2 Venture Partners."
Food delivery robots hit Canadian sidewalks, but many challenges delay mass adoption - "When customers in downtown Vancouver placed orders with Pizza Hut in September, many of the pies landed on their doorsteps without a courier in sight. Instead, diners were met by Angie, Hugo or Raja -- autonomous robots resembling a cooler on four wheels with eyelike lights."
Adobe Stock will sell images created with the help of artificial intelligence - "Adobe has announced today it is now accepting images created with the help of generative artificial intelligence applications like Stable Diffusion and DALL-E, opening up the sale of these images at Adobe Stock."
Meta is expanding its use of AI face scanning to verify users’ age on Facebook Dating - "Facebook is testing ways for people to verify their age when using the platform’s dating app, including using an AI face scanning tool."
Amazon tests machine learning software to analyze satellite images from space - "Artificial intelligence in space can revolutionize Earth-observation."
‘We get to be their visual storytelling partners’: How artificial intelligence is transforming the film and entertainment industry - "While Bredow said he expected machine learning and A.I. to be extremely transformative, it’s gone beyond that."
Intuit CEO Sasan Goodarzi says A.I. is turning the TurboTax parent into more than a 'transactional' platform - "Intuit CEO Sasan Goodarzi is quick to explain why he’s placed artificial intelligence (A.I.) at the center of his company"
AI text generation is moving mainstream with Canva’s Magic Write - "Today, Canva announces Magic Write, a text-generation tool that can generate everything from ideas for blog posts to a cover letter. But AI text is quietly — and probably inevitably — moving into more products you’ll use on a regular basis. AI art is already there."
Runway Announces $50M Series C Led by Felicis - "BusinessRunway Announces $50M Series C Led by Felicis"
Howie Mandel gets a digital twin from DeepBrain AI - "Howie Mandel is stepping into the metaverse. DeepBrain AI has created a pretty realistic AI version of comedian and actor Mandel."
Xiaomi’s Humanoid Drummer Beats Expectations - "Solving drum-playing helped quest for whole-body control"
Toastmasters adopts AI-powered speech analytics technology from Seattle startup Yoodli - "Can artificial intelligence strengthen human connections by improving public speaking? That’s the ambition of a new partnership between Toastmasters International and Yoodli."
Uber and Motional Bring Robotaxi Service to Las Vegas - "Uber has joined forces with Motional to bring robotaxis to the streets of Las Vegas. Soon, they may be everywhere. The ride-hailing company announced the launch with a press release Wednesday."
Concerns
How the Collapse of Sam Bankman-Fried’s Crypto Empire Has Disrupted A.I. - "Mr. Bankman-Fried and his colleagues spent more than $530 million to battle what they saw as the dangers of artificial intelligence. Now those efforts are reeling."
Your selfies are helping AI learn. You did not consent to this. - "What are the rules of the road for the A.I. age, where anything you share online might train a computer system that puts an innocent person in jail?"
AI art generator DreamUp illustrates the collision between artistic consent and AI datasets - "Should it be an issue that countless data points of artwork are used to train AI that spits out art in under a minute?"
Clip Studio Paint cancels AI image generator amidst backlash from users - "Japanese illustration app Clip Studio Paint has canceled the rollout of a recently announced AI image generator tool after backlash from its users."
AI-generated answers temporarily banned on coding Q&A site Stack Overflow - "Stack Overflow, the go-to question-and-answer site for coders and programmers, has temporarily banned users from sharing responses generated by AI chatbot ChatGPT."
Uber’s facial recognition is locking Indian drivers out of their accounts - "Some people are finding their accounts permanently blocked"
ChatGPT shows promise of using AI to write malware - "For even the most skilled hackers, it can take at least an hour to write a script to exploit a software vulnerability and infiltrate their target. Soon, a machine may be able to do it in mere seconds."
Google SVP says the search giant has to be ‘thoughtful’ about the range of nefarious uses accompanying artificial intelligence - "As A.I. becomes more and more advanced, companies working to explore this world have to tread deliberately and carefully."
The inherent misogyny of AI portraits – Amelia Earhart rendered naked on a bed - "The hugely popular Lensa AI app creates portraits based on image prompts, but users say it sexualizes them unnecessarily"
The Magic Avatar you paid $3.99 for is probably stolen, artists say - "Artists say Lensa's Magic Avatars are built by AI that uses techniques and styles that were taken from artists without consent or compensation."
Analysis
On the Emergence of General Computation from Artificial Intelligence - "we will soon witness the (re-)emergence of general computation from artificial intelligence."
AI Homework - "In other words, the role of the human in terms of AI is not to be the interrogator, but rather the editor."
ChatGPT Is Dumber Than You Think - "Treat it like a toy, not a tool. As a critic of technology, I must say that the enthusiasm for ChatGPT, a large-language model trained by OpenAI, is misplaced."
The Truth About AI Getting "Creative" - "Let's talk about AI Art, Lensa, ChatGPT, and why it's all deeper than you think."
Policy
The TSA's facial recognition technology, which is currently being used at 16 major domestic airports, may go nationwide next year - "The TSA may expand the pilot program of its facial recognition identification system — currently being used in 16 domestic airports across the United States — to include airports nationwide as early as next year."
Explainers
Building Airbnb Categories with ML and Human-in-the-Loop - "we sought to inspire the traveler to book unique stays in places they might not think to search for"
Fun
Using ChatGPT As a Co-Founder - "ChatGPT has been blowing people’s minds. I thought I’d give it the challenge of helping to build a startup. Let’s see how it does. I’m creating a startup that will allow businesses to easily ask their customers to upload files."
No Linux? No problem. Just get AI to hallucinate it for you - "Over the weekend, experimenters discovered that OpenAI's new chatbot, ChatGPT, can hallucinate simulations of Linux shells and role-play dialing into a bulletin board system (BBS)."
The Brilliance and Weirdness of ChatGPT - "A new chatbot from OpenAI is inspiring awe, fear, stunts and attempts to circumvent its guardrails."
Copyright © 2022 Skynet Today, All rights reserved.
Last Week in AI #197: AI for Drug Discovery, Deep Reinforcement Learning for Atom Manipulation, SF rolls back plan for killer police robots, and more!
I always appreciate these summaries. You include the good (more examples of positive AI developments than normally seen these days), the bad, and the ugly. All of which are equally valuable.
The SF police robot scenario deserves a more comprehensive review. At least to me, it’s not a binary answer. The forthcoming dialogue will be invaluable in shaping national policies, processes, and procedures.