Welcome to a special edition of Eye on A.I.
Consumer tech giants like Facebook-parent Meta and Twitter may steer artificial intelligence to be more explainable and to rely on less data.
Take Twitter, for instance, which just accepted a $44 billion takeover bid by Tesla chief Elon Musk. One of Musk’s plans for the company after he takes control—likely in around six months—is to open source its algorithm so that the public can check whether it covertly manipulates what users see and post on the service.
But there’s several problems with Musk’s proposal. A Twitter technologist told Fortune that Twitter isn’t comprised of “just one algorithm,” but is instead “a web of interconnected systems that work together to show tweets to each user.” And even if people were to inspect the company’s various algorithms and data pipelines, it’s unclear they’d even be able to make sense of the technology.
“Even if you could open source one of the machine learning models, you likely wouldn’t be able to understand it without also understanding all of the inputs and outputs, and how they are transformed along the way,” the Twitter employee said.
Essentially, what Musk wants to do is solve one of A.I.’s most complicated issues: Explaining how deep-learning systems actually make decisions.
Musk has been vague about how he would make Twitter’s “algorithm” open source. But the core of his idea is predicated on making Twitter’s underlying A.I. systems decipherable—a huge challenge with no practical solution, as my colleague Jeremy Kahn has previously reported.
Indeed, the feat of making Twitter’s A.I. systems understandable could be as big of a challenge as Musk’s other endeavors, such as creating cutting-edge implantable brain computing chips or sending people to Mars.
Meanwhile, Meta is embarking on its own A.I. grand challenge: Creating A.I.-powered tools that help display relevant ads to users without requiring truckloads of personal data. Meta CEO Mark Zuckerberg and chief operating officer Sheryl Sandberg discussed the need multiple times during the company’s latest earnings call.
Apple’s recent privacy changes to its iOS mobile software has severely hampered Meta’s ability to provide businesses with tools that can accurately targets ads. Sandberg told analysts on Wednesday that Meta is “developing privacy enhancing technologies, which will help us minimize the amount of personal information we process, while still allowing us to show people relevant ads and measure performance for advertisers.”
This is no easy task, and it will require Meta to invest heavily in A.I. for the rest of the year, at the least. But if Meta is able to create A.I. tools that require less data than older tools, the results could be profound for many other businesses that have long been operating under the A.I. mantra that the more data, the better the machine learning model.
Jonathan Vanian
@JonathanVanian
jonathan.vanian@fortune.com
A.I. IN THE NEWS
More Adept A.I. Two creators of a powerful kind of neural network software called transformers have founded business startup Adept, which raised $65 million, TechCrunch reported. Adept’s chief scientist Ashish Vaswani and chief technology officer Niki Parmar both hail from Alphabet’s A.I. research division Google Brain, where their work on transformers helped teach computers to understand and automatically generate more human-like language. Some of the most powerful language systems like OpenAI’s GPT-3 and Google’s BERT software are powered by transformers. Adept is now attempting to use transformers to power cutting-edge software assistants that can learn to do business-related tasks like automatically generate compliance reports or draw blueprint images when asked, the report said.
Animal sickness. Georgetown University researchers are using machine learning to predict which animal viruses infect humans, The New York Timesreported. The researchers told the publication that the machine learning software could also help identify which animals may be more likely to “harbor dangerous viruses we don’t yet know about.” From the article: “It feels like you have a new set of eyes,” said Barbara Han, a disease ecologist at the Cary Institute of Ecosystem Studies in Millbrook, N.Y., who collaborates with Dr. Carlson. “You just can’t see in as many dimensions as the model can.”
The A.I. knows that you will miss your meeting. Researchers at Boston Children’s Hospital are exploring how machine learning can be used to predict which patients may miss their medical appointments, the health news publication Stat reported. The goal is for the technology to help healthcare institutions “target interventions to the patients at highest risk of missing their appointments and offer them whatever help they need making it in,” the report said. Still, the project is only research and Boston Children’s Hospital did not say when or if the technology will be more widely used.