New Pope calls AI a “challenge for humanity”

Leo XIV, the “Chicago Pope”, and the first American pope in the Catholic Church’s 2,000-year history, has made AI a subject of concern from his very first days. His choice of name, Leo, harks back to the previous Pope Leo XIII, who in 1891 wrote an encyclical on the impact of the Industrial Revolution on humanity. In the new pope’s first major address, he declared that AI “poses new challenges for the defense of human dignity, justice, and labor.” As a member of the Augustinian order, he has been steeped in the order’s values of community and service. His concern for social justice has caused him to be called the “Woke Pope” by some, and his election sends a strong message that the Church establishment continues to support the social justice and inclusivity policies of the recently deceased Pope Francis. Expect the new pope to speak out repeatedly on issues of AI disruption of labor, and any use of AI for manipulation or depersonalization.

Pope Leo’s first major address to the College of Cardinals

Saudi Arabia launches AI Startup; Saudis and UAE may get Nvidia chips

Saudi Arabia has announced that it is launching a major new AI startup, called Humain, backed by the kingdom’s $930 billion sovereign wealth fund. The company will be chaired by Crown Prince Mohammed bin Salman (known colloquially as MBS), and is a major plank in the kingdom’s Vision 2030 roadmap for making the Saudi economy less dependent on oil. MBS has been under a cloud in recent years on account of human rights abuses, and the horrific murder and dismemberment of a journalist, but a trillion-dollar investment fund is just too tempting a prize to discourage American companies for too long. Elon, Zuckerberg, and Sam Altman are all in the country now for a US-Saudi investment forum. MBS announced a plan to invest $600 billion in US companies over the next 4 years, and President Trump in his current visit to the kingdom is brokering deals to allow Nvidia chips to flow to the new Humain AI startup, as well as to the United Arab Emirates’ more established G42 AI startup. The next stage of AI advances is likely to be fueled significantly by Middle East petrodollars.

Saudi Arabia launches AI startup, and Saudis and the UAE may get Nvidia chips

Netflix rolls out AI search so you can find a movie to watch

Netflix has partnered with OpenAI to bring AI-enabled natural language search to the task of finding a movie to watch. Instead of having to input keywords, users will be able ask for movies any way they like, such as “a scary movie, but not too scary, and also funny.” The AI search will take into account whatever it knows about your watch history, but allow you to add nuances to your request that Netflix’s current search function is unable to process. The AI search tool was first piloted in Australia and New Zealand, and is now being released as an opt-in beta product in the US.

Netflix rolls out AI search to find a movie to watch

Perplexity raising $500 million at a valuation of $14 billion

AI search startup Perplexity continues to grow fast, making its valuation skyrocket. It is now in the final stages of talks to raise an additional $500 million at a price that values the company at $14 billion (rumors are, Perplexity wanted to be priced at $18 billion. There’s just no satisfying some people.) Just 6 months ago, the company raised $500 million at a valuation of $9 billion; a year ago, it raised $250 million at a valuation of $3 billion. Skyrocket, indeed.

/

Perplexity raising $500 million at a valuation of $14 billion

Meta-analysis of 51 studies shows students learn more with AI

Use of AI in education has been controversial, with widespread hand-wringing about how students may use AI to avoid learning. Last week, 2 Chinese academics published an article in Nature reporting on a meta-analysis of 51 high-quality studies in the literature that assessed the impact of ChatGPT on learning in primary, secondary, and college education. Surprising to many of the nay-sayers, the analysis indicated that use of ChatGPT had a large positive effect on students’ learning performance, with smaller but still significant positive effects on learning perception and on critical thinking. The authors argue that “ChatGPT should be flexibly integrated into teaching as an intelligent tutor, learning partner, and educational tool.”

Nature article on the effect of ChatGPT on learning

Carnegie Mellon releases LegoGPT, text-to-Lego construction planner

AI is increasingly being used to solve real-world problems in business, science, engineering, and a myriad of other fields. Now the big-brained scientists at Carnegie Mellon University have used AI to solve a particularly vexing challenge - how to design novel Lego construction projects that both look good and don’t fall apart. In their magnanimity, they have released LegoGPT, an open-source AI model that designs a stable, buildable Lego construction based on a simple text prompt. The model will show all pieces required, and animates a step-by-step construction plan that can be used by humans or even by industrial assembly robots. Check out the link below to see the range of what can be produced, and animations showing the block-by-block construction.

Carnegie Mellon releases LegoGPT, a text-to-Lego construction planner

AI refurbishes and expands Candy Crush’s 18,700+ levels

The venerable time-waster, Candy Crush, a wildly popular smartphone game that turned 13 years old last month, has kept its 180 million loyal users coming back by continually expanding game levels to keep the game fresh. Over the years, levels have expanded to a staggering 18,700-plus. Now the game is using AI code generation to drive a top-to-bottom renovation of existing levels, and a significant expansion in the number of levels. The hope is to maintain, and if possible increase its whopping $1 billion dollars in annual revenue. Since that represents only about $5 per user per year, even a small increase in revenue per user can drive outsized returns that fall directly to the bottom line.

Candy Crush is using AI to renovate and expand its 18,700+ levels

Beijing to host world’s first Humanoid Robot Games

Beijing will host the world’s first Humanoid Robot Games from August 15-17, an Olympics-style sports contest that will feature humanoid robots from multiple manufacturers competing in various sporting events. The Games will feature 19 different events, including floor exercises, soccer, and dance. The Games are part of a concerted push by China to showcase and improve its robotics technology, with the aim of becoming a world leader in the field. We reported on China’s recent humanoid robot half-marathon footrace competition. China is also hosting a World Robot Conference just prior to the Games, and says that it expects 200 companies to attendc and show their latest models.

Beijing to host world’s first Humanoid Robot Games

Quick Takes:

China uses drones to create rain in drought area

Chinese government scientists have recently released results of a weather modification experiment, in which a fleet of drones released silver iodide particles into clouds above the drought-stricken Xinjiang area of the country, increasing rainfall by about 4%, or 18.5 million gallons. The hope is to be able to scale this small proof-of-concept up to a level that would make a meaningful contribution to local agriculture. (link)

France aims for combat-ready robots by 2028

The French armed forces have been holding annual competitions of military robots for several years, with the original goal of having combat-ready robots by 2040. The drone war in Ukraine has enormously accelerated development in this field, and has provided increased motivation for EU countries to put effective solutions on the battlefield in order to deter and counter Russian aggression. The new timetable is to have combat robots in service within 3 years, spokespeople say. (link)

Nvidia simulated 10 years of training for humanoid robots in 2 hours

Premier AI chipmaker Nvidia has a robust robotics research program, seeing this sector as a major driver of future chip sales. The company recently demoed a new method of training robots on complex tasks in a simulated environment. The simulated environment can be run thousands of times faster than reality, allowing 10 years of training to be delivered to the robot in 2 hours. (link)

OpenAI says newest models equal physician performance on benchmark

OpenAI has released HealthBench, a new benchmark test for AI models in the field of health care. The benchmark consists of 5,000 realistic simulated conversations between a human and an AI, each of which sets up a clinical scenario which calls upon the AI to make a diagnosis and outline a plan of action. The test was devised with the input of 262 physicians practicing in 60 different countries. The test was administered to several different OpenAI models, some older ones from last year as well as the newest ones, and to a panel of physician experts, some of whom were given access to an AI model to assist them, and some who were not. The major outcomes were: 1) the newer models performed significantly better than the older ones, scoring about 80% higher. 2) Physicians who used the older models to help craft a response scored approximately 14% higher than the unaided older model - that is, they improved on the model’s responses. 3) Physicians were unable to improve the score of the newer models - evidence, according to OpenAI, that the models performed as well as the physicians. Now, this is a new benchmark, and its methodology will need to be rigorously tested, but this is another indication that AI is rapidly gaining the ability to process clinical information at a very high level.

OpenAI says its newest models equal physician performance on benchmark

MGB AI predicts cancer survival from a face photograph

An international team led by Massachusetts General Brigham has developed FaceAge, an AI system that estimates biological age from a photograph of a person’s face. This model has been found to improve estimates of survival times of cancer patients as compared to use of chronological age, and as compared to physicians’ subjective impression of the patient’s biological age.

MGB AI predicts cancer survival from a face photograph

Quick Takes:

FDA announces aggressive rollout of AI

Last week FDA Commissioner Martin Makary MD MPH announced completion of a highly successful pilot program for using AI to assist scientific reviews of new drugs. Based on this success, the FDA is planning an aggressive rollout of AI in the scientific review process, with the goal of speeding decisions on safety and efficacy, in order to shorten the time for approval of promising new drugs. (link)

Oura ring fitness wearable pairs with continuous glucose monitor

Oura makes a fitness wearable in the shape of a ring. Despite the miniature form factor, the ring seems to track physiologic variables such as heartrate and steps more or less comparably to the more familiar Fitbit or Apple Watch. Now the company has announced an alliance with Dexcom, the dominant maker of continuous glucose monitors. This feature is being paired with a separate Meals feature in Oura’s smartphone app, in which the user can take a picture of each meal, and Oura’s AI in the cloud will recognize the food components and portions, estimate calories and proportion of fat, carbohydrate and protein, and analyze the effect of that meal on the user’s blood glucose as measured by the Dexcom CGM. This sort of multi-level AI analysis of personal health is likely to grow substantially in the future. (link)

Epic rolls out AI-enabled scheduling via text messaging

Electronic medical record software giant Epic is tackling a major source of patient dissatisfaction - how hard it is to set up a doctor’s appointment. In a world where consumers are able to reserve a table at a restaurant or order delivery of almost any physical object at the click of a button, having to call a doctor’s office during work hours to speak to a human being feels almost medieval. Now Epic is using AI to enable appointment scheduling through text messaging. An AI chatbot handles both the text conversation and the actual appointment scheduling in the Epic system. (link)