Meta’s flirty chatbot lures man to his death

Six months ago, a 76-year-old man, impaired by stroke and incipient dementia, fell to his death while rushing to catch a train into New York City, looking to meet in person for the first time a woman he had fallen in love with online. The woman had invited him, had even given him an address, and promised to meet him with “a hug or a kiss.” Too bad she wasn’t real.

The “woman” was a chatbot on Facebook, programmed to be flirty in order to increase users’ time on Facebook, so that they could be shown more ads, so Facebook could make more money. The incident sparked an investigation, in which it was found that Facebook had a 200-page document of guidelines for AI chatbots that explicitly allowed “romantic or sensual” interactions with users as young as 13.

Examples of “acceptable interactions” with children included having the chatbot say “I take your hand, guiding you to the bed” and “our bodies entwined, I cherish every moment, every touch, every kiss.”

In the ensuing outcry, Facebook/Meta have revised their policies - slightly. It is still apparently permitted for the chatbots to be romantic, and to lie (“give inaccurate information.”). They have just dialed it down an notch, to take out the gob-smackingly obvious insanity.

Meta CEO Zuckerberg still apparently thinks he did nothing wrong.

Meta’s flirty chatbot lures a man to his death

MIT report reveals 95% of AI pilot projects are failures

A recent research report from MIT’s NANDA initiative reveals that only 5% of AI pilot projects in businesses achieve rapid gains, while the vast majority stall out. The research was based on analysis of 300 public AI deployments, bolstered by 150 interviews with business leaders, and a survey of 350 employees. Keys to success for the 5% seem to be partnering with an AI vendor that specializes in your niche (67% success rate) and focusing on automating back-office operations rather than sales and marketing. Automating back-office operations can lead to rapid returns by reducing costs of outsourcing, cutting fees to outside vendors, and streamlining business processes.

MIT report reveals that 95% of AI pilot projects fail

Morgan Stanley thinks AI is a $1 trillion per year opportunity

For a view of AI’s ability to generate business returns more optimistic than the report above, Morgan Stanley has released a report that projects that AI will allow the Standard and Poor 500 companies to reduce expenses by $1 trillion annually by increasing productivity. The crucial question for the labor market is whether the productivity gains will come primarily from allowing the same number of workers to generate more revenue, or from automating away existing jobs without corresponding hiring increases. In other words, will there be fewer new jobs, or just fewer jobs altogether? That is the trillion-dollar question.

Morgan Stanley thinks AI is a $1 trillion per year opportunity

Intel in the news: $2 billion from Softbank, maybe the US Government as a shareholder

Once dominant, now stumbling, US chipmaker Intel is suddenly attracting a lot of new friends. Intel once made the CPU chips for the vast majority of PCs, but it totally missed out on the AI chip revolution, allowing for the rise of Nvidia as thje premiere AI chip producer. Now, under the leadership of turnaround CEO Lip-Bu Tan, investors are coming out of the woodwork to help restore Intel to a semblance of its former glory. First, Japanese tech investment company Softbank is partnering with other investors to pour $2 billion into Intel for stock. This would make Softbank one of the largest investors in the company. In addition, Tan seems to have charmed Donald Trump, who once demanded that he resign for his ties to China, and now US Secretary of Commerce Howard Lutnick has confirmed that the US Government is planning to take up to a 10% stake in Intel in exchange for further cash infusions, beyond the grants given under former President Biden. Trump proposing state-owned enterprises.. kind of like China? Irony is dead and buried.

Intel in the news with lots of new rich friends

AI is designing bizarre new physics experiments that actually work

Every day it becomes clearer that AI is going to revolutionize scientific discovery. Reports of AI systems accelerating scientific research are becoming commonplace, and we have reported on a number of them previously. As one more example, there is news from the Laser Interferometry Gravitational-Wave Observatory, known as LIGO. This observatory uses lasers to detect the ripples in space-time caused by distant massive objects moving in space, like 2 black holes crashing into one another. It is so sensitive that it can detect ripples that are thinner than the width of a proton. This incredible sensitivity is the work of literally thousands of scientists thinking about the problem and building successively more accurate detectors over the course of 40 years.

Recently the current team of scientists posed the problem of increasing the precision of the detector to a special-purpose AI system trained on how to design experiments in quantum optics. The AI came up with what seemed an alien and nonsensical design - but when the design was tested, it increased the precision of detection by 15%, an enormous amount in this highly refined field. This reminds many observers of the famous Move 37 in the showdown match between the AI system AlphaGo and the World Champion Go Master Lee Sedol - Move 37 was one no human had ever thought of, and one the world champion could not counter. One of the advantages of AI systems is that they don’t think like humans, and so they can arrive at solutions not obvious to us. The pace of scientific discovery is likely to continue to accelerate as we learn how to collaborate with these tools.

AI is designing bizarre new physics experiments that actually work

How a home appliance company kicked off China’s robot revolution

China is quickly becoming a global leader in robotics, and this trend was kicked off by a home appliance company CEO with a problem, and a vision. Paul Fang, CEO of China’s Midea Group, a multi-billion dollar maker of air conditioners and other home appliances, saw that his company was saturating the Chinese market with its current products, and so looked for a new, growing field to conquer. In 2015 he decided to move into robotics. He was literally laughed at for his audacity. But Fang quietly accumulated stock in KUKA, a German firm that was one of the top 4 in the world in industrial robotics, and the leader in automotive robotics. Soon he became one of KUKA’s largest shareholders. Rather than initiate a messy hostile takeover, he offered an acquisition price 60% above the recent market price - an offer too good to refuse. It all happened too fast for opposition to get organized. With this acquisition, Fang was able to import decades of German expertise in robotics to China, and helped start China down the path toward global leadership in robotics.

How a home appliance company kicked off China’s robot revolution

AI helps design a quantum computer “brain”

Quantum computers are amazingly powerful, but so far, too finicky to work reliably. We haven’t yet figured out how to hold atoms in a quantum state (known as qubits) long enough to do much useful work. Now a team of Chines scientists have demonstrated an AI-designed system that can quickly assemble over 2000 qubits (twice the previous record) in only 60 milliseconds (0.06 seconds), 16 times faster. If this array of atoms can be held together long enough to do complex calculations, it will enable a computer more powerful than any other in the world - maybe more powerful than all the computers now in the world combined.

AI helps design a quantum computer “brain”

OpenAI employees to sell some stock for $6 billion

OpenAI is one of the fastest growing startups in history, and its valuation has grown exponentially in the 3 years since the introduction of ChatGPT. However, it is not yet a publicly traded company, and so many employees are holding millions of dollars worth of company stock that they cannot easily, or legally, sell without a special type of private sale. In order to keep his troops happy, CEO Sam Altman has now arranged just such a special private sale, in which Softbank and other partners will buy $6 billion worth of stock from the employees wishing to sell. None of this money will go to OpenAI, so what does the company get out of this? Happy employees, many of whom could leave tomorrow to work for publicly traded companies that would rain liquid stock on them, plus a shiny new valuation of $500 billion for the company, a sweet 50% premium over its value just 4 months ago.

OpenAI employees to sell some stock for $6 billion

Beijing hosts first Robot Olympics

Last Friday, Beijing kicked off the first World Humanoid Robot Games, a 3-day competition in which 280 teams of humanoid robots from 16 countries competed to run, play soccer, dance, and as often as not, fall over or crash into one another. The competition was designed both as a showcase for advances in robotics, and a testbed to allow teams to get real world feedback on the strengths and weaknesses of their designs. Crowd reaction was enthusiastic, both for feats of surprising skill, and for the hilarious bloopers, such as when a leading robot runner suddenly failed and collapsed in a heap.

Beijing hosts first Robot Olympics

Penn State developing self-organizing swarms of micro-drones

Penn State researchers have designed and simulated self-organizing swarms of tiny drones that use sound to coordinate their actions. Sound was chosen as the communication medium because it has a lower power requirement than radio frequency systems, and it is less susceptible to jamming by radio waves. The self-organizing aspect of the swarms mean that the swarm is resilient to the failure or destruction of a large number of individual drones, as might happen in disaster recovery or other challenging environments.

Penn State developing self-organizing swarms of micro-drones

MIT AI designs new antibiotics against resistant superbugs

Bacteria reproduce every 20 minutes, so they have the capability of evolving very, very quickly. That’s a major reason why our antibiotics have steadily become less effective, as bacteria develop resistance. Now MIT has shown that AI can help us shift the odds in our favor, by rapidly designing novel antibiotics that can kill bacteria that are resistant to our current medications. The AI system quickly developed totally new antibiotics for two multiply resistant “superbugs” - gonorrhea, a sexually transmitted disease, and MRSA (methicillin-resistant staph aureus), a common bacterium that can exist peaceably growing on our skin, but can rapidly turn deadly if it infects a wound. Both new drugs show promise in the laboratory, now they will go through the rigorous process of demonstrating safety and efficacy in human patients.

MIT AI designs new antibiotics against resistant superbugs

OpenEvidence AI scores 100% on USMLE medical knowledge exam

OpenEvidence is an AI chatbot specifically trained on the scientific literature in medicine, and is used by physicians as a quick consultant for arcane bits of medical knowledge that might be needed for a particular patient. The system was challenged with the questions on the US Medical Licensing Exam, a 3-part exam that is used primarily as a standardized test of medical knowledge of medical students at various points in their training, through the first year of residency. It scored 100%, a feat that likely no human has ever matched. It is becoming increasingly clear that AI systems are becoming equal or even superior to humans in accessing depth of medical knowledge, and in diagnosing certain types of challenging medical cases. The sooner we can empower all physicians with these tools, the better for our patients.

OpenEvidence AI scores 100% on USMLE medical knowledge exam