This story first appeared in The Algorithm (our weekly newsletter about AI). Sign up to receive stories like these in your inbox first.
It's horrible out there. Each of the top tech companies in the world has announced mass layoffs. Alphabet, a tech company that employs 12,000 people, announced its plans to lay off last week. Amazon, Meta and Microsoft have all announced layoffs that are causing havoc. These layoffs affect not only AI researchers, but entire AI teams.
It was sad to see the story over the weekend of how Googlers in the US learned about the abrupt cull. Dan Russell, a researcher scientist who has been working on Google Search for more than 17 years, described how he went to work at 4 AM and found that his entry badge did not work.
Economists believe the US economy could enter recession this year, despite a uncertain global economic outlook. The squeeze is starting to hit big tech companies.
Economic downturns in the past have prevented funding for AI research. These are known as "AI winters" but this time, we are seeing something completely different. AI research is still very hot. It's making huge leaps in progress, even though tech companies are tightening their belts.
Big Tech actually believes that AI will give them an advantage.
AI research has been in and out fashion since its inception in the 1950s. AI research has been prone to hype cycles of unrealistic expectations and subsequent failures. Peter Stone, a professor of computer science at the University of Texas at Austin, was a former employee of AT&T Bell Labs (now Nokia Bell Labs), who worked on AI until 2002.
Bell Labs was a hot spot for innovation for decades. Its researchers won numerous Turing Awards and Nobel Prizes, including Yoshua Bengio, Yann LeCun and Geoffrey Hinton. As management began pushing for immediate returns on technological improvements, the lab's resources were reduced and it ultimately failed to capitalize on the internet revolution of early 2000s. Jon Gertner writes about this in his book The Idea Factory: Bell Labs & the Great Age of American Innovation.
Stone says that the previous downturns occurred after the most advanced AI techniques of the day failed in their progress, were unstable and hard to run. The funding provided by the US and UK governments for AI research was quickly cut.
AI research today is experiencing its "main character". Despite the economic downturn, AI research remains exciting. Michael Wooldridge, a professor of computer science at Oxford and the author of A Brief Historical of AI, says that "we are still seeing regular rollouts of system which are pushing back to the frontiers of AI can do."
This is quite a contrast to the reputation of the field in the 1990s when Wooldridge was finishing up his PhD. AI was still considered a fringe pursuit. The wider tech sector saw it in the same way that established medicine sees homeopathy.
The neural networks are the engine of today's AI research boom. They were created in 1980 and have been instrumental in simulating patterns within the human brain. The technology was not powerful enough to run it back then. The technique is now possible because of the abundance of data available and the power of computers.
Every few months, new breakthroughs like ChatGPT, a chatbot, and Stable Diffusion (a text-to-image model) seem to be made. Stone says that technologies like ChatGPT have not been fully explored and that both academia and industry are still trying to figure out how they can be of use.
Instead a full-blown AI winter we will likely see a decrease in funding for long-term AI research as well as more pressure to make money with the technology, according to Wooldridge. He says that corporate lab researchers will have to demonstrate their ability to integrate their research into products, and make money.
This is already happening. Google is looking to revamp Search using its own AI research, in light of OpenAI's ChatGPT's success.
Stone sees parallels between what happened at Bell Labs and Stone's. He believes that if Big Tech's AI labs focus too much on short-term product development and turn away from deeper, more long-term research, then these large labs may lose their grip on innovation.
This is not necessarily a negative thing. There are many smart people out there looking for work right now. As crypto crashes out, venture capitalists are searching for startups to invest in. Generative AI has proven that the technology can be turned into products.
The AI sector has a rare opportunity to explore the possibilities of new technology. It's an exciting prospect, despite all the negativity surrounding layoffs.
We've created a new series of reports based on MIT Technology Review's 10 Breakthrough Technologies. The first, which will be released later in the week, is about how engineering and industrial design firms use generative AI. Register to be notified when the book is available.
Deeper Learning
AI brings the internet to submerged Roman ruin
Baiae, a magnificent resort town in the Italian peninsula, was established over 2,000 years ago. Rich statesmen built luxurious villas in the natural springs to enjoy their thermal pools and heated spas. Over the centuries, volcanic activity submerged the Roman nobility's playground, leaving half of it below the Mediterranean. It is now a protected marine area that must be monitored for environmental factors and divers. However, communication underwater can be extremely difficult.
Underwater: Italian researchers believe they have found a new way for the internet to be brought underwater. AI and algorithms adjust network protocols to the sea conditions, allowing the signal to travel as far as two kilometers. This could allow researchers to better understand the impacts of climate change on marine environments, and help them monitor underwater volcanoes. Although AI research can seem abstract, this is an example of how technology can be used. Manuela Callari has more information.
Bits and Bytes
OpenAI used low-paid Kenyan workers in order to make ChatGPT more toxicOpenAI used Sama, a Kenyan company, to train its AI system ChatGPT to create safer content. Low-paid workers had to sift through a lot of violent and graphic content, including incest, child sexual abuse, bestiality and murder. This is a reminder of the horrible work that humans do behind the scenes in order to keep AI systems safe. (Time)
CNET's AI-powered, SEO money machine
CNET, a tech news site, has begun using ChatGPT for news articles. The site has had to correct factual errors in articles since they were published, which is not surprising. The Verge examined why CNET chose to use AI to create stories. It's a tragic tale about what happens when journalism and private equity collide. (The Verge)
China could be a model for deepfake regulations
Deepfakes have been resisted by governments because they fear that it may restrict free speech. China's government isn’t as concerned about this risk and believes it has a solution. For example, deepfakes must have consent from the subject and bear watermarks. Other countries will also be paying attention and taking notes. (The New York Times).
Nick Cave believes a ChatGPT song written in his style is a suck
Perfection. No comments. Chef's kiss. (The Guardian)
————————————————————————————————————————————————————————————
By: Melissa Heikkilä
Title: The economy is down, but AI is hot. Where do we go from here?
Sourced From: www.technologyreview.com/2023/01/24/1067232/the-economy-is-down-but-ai-is-hot-where-do-we-go-from-here/
Published Date: Tue, 24 Jan 2023 10:48:30 +0000
Leave a Reply