Every year, machines become more intelligent, yet artificial intelligence has failed to live up to the hype created by some of the world’s top technological corporations.
AI can succeed at specific restricted tasks, such as chess, but it struggles to master several ones. A seven-year-old, for example, has significantly more intelligence than any of today’s AI systems.
According to Edward Grefenstette, a research scientist at Meta AI, previously Facebook AI Research, “AI algorithms are good at approaching individual tasks, or tasks with a minimal degree of variety.”
“However, the real world contains tremendous potential for change,” he continued, “a dynamic that we are lousy at capturing inside our training algorithms, resulting in fragile intelligence.”
According to Grefenstette, AI researchers have begun to demonstrate that there are ways to efficiently adapt AI training methods to changing settings or tasks, resulting in more robust agents. This year, he expects, more commercial and scientific implementations of such technologies will result in “noticeable leaps.”
While artificial intelligence still has a long way to go before it approaches human intelligence, that hasn’t stopped companies like Google, Facebook (Meta), and Amazon from investing billions of dollars in hiring talented AI researchers who can potentially improve everything from search engines to voice assistants to aspects of the so-called “metaverse.”
Claims about the effectiveness and reality of AI in spaces that are now being labeled as the metaverse will become more commonplace in 2022, according to anthropologist Beth Singler, who studies AI and robots at the University of Cambridge, as more money is invested in the area and the public begins to recognize the “metaverse” as a term and a concept.
Singler also cautioned that in 2022, there may be “too little conversation” about the metaverse’s impact on people’s “identities, communities, and rights.”
Gary Marcus, a scientist who sold an AI start-up to Uber and is now executive chairman of another company called Robust AI, told CNBC that the most significant AI innovation in 2022 would likely be one that the public doesn’t notice right away.
“It can take years to go from lab discovery to practice,” he said, adding that deep learning still has a long way to go. Deep learning is a branch of AI that aims to learn how to recognize complicated patterns in data by simulating the activity of layers of neurons in the brain.
“I predict that progress on this problem will be transformational this year, towards what I dubbed hybrid systems,” Marcus continued, “but that it will be a few years before we see substantial returns.” “The first pharmaceutical in which AI had a significant role in the development process will most likely be seen this year or next.”
The next steps for DeepMind
DeepMind, an Alphabet-owned research group based in London, has made one of the most significant AI achievements in recent years.
The company has successfully developed AI software that can properly anticipate the structure that proteins will fold into in a few of days, resolving a 50-year-old “grand issue” that could lead to better disease understanding and treatment discovery.
DeepMind will likely pursue additional major science topics in 2022, according to Neil Lawrence, a professor of machine learning at the University of Cambridge, who told CNBC.
In 2022, language models – AI systems that can write convincing text, speak with humans, answer questions, and more — will also advance.
OpenAI’s GPT-3 is the most well-known language model, although DeepMind claimed in December that their new “RETRO” language model can outperform others 25 times its size.
Big Tech, according to Catherine Breslin, a machine learning expert who previously worked on Amazon Alexa, will sprint toward greater and larger language models in the coming year.
Breslin, who now runs AI consultant business Kingfisher Labs, told CNBC that instead of treating vision, speech, and language as independent jobs, there will be a drive toward models that mix them.
According to CNBC, a new breed of startups will likely employ language models to predict the most successful RNA (ribonucleic acid) sequences, according to Nathan Benaich, a venture capitalist with Air Street Capital and co-author of the annual State of AI report.
“We saw the impact of RNA technologies last year when innovative covid vaccinations, many of which were built on this technology, ended nationwide lockdowns,” he said. “I expect a new crop of AI-first RNA therapeutic firms will emerge this year. These new companies could drastically reduce the time it takes to develop new medications and vaccines by using language models to forecast the most effective RNA sequences to target a disease of interest.”
Concerns about ethics
While many breakthroughs may be on the horizon, there are serious ethical concerns with AI, which can be highly discriminatory and biased when trained on particular datasets. Artificial intelligence is also being utilized to power autonomous weapons and create fake pornography.
In 2022, according to Verena Rieser, a professor of conversational AI at Heriot-Watt University in Edinburgh, there will be a greater focus on ethical issues surrounding AI.
“I’m not sure if AI will be able to do much ‘new’ stuff by the end of 2022,” she remarked, “but maybe it will do it better,” she said, implying that it will be fairer, less biased, and more inclusive.
According to Samim Winiger, an independent AI researcher who previously worked for a Big Tech business, there will be exposes about the use of machine learning models in financial markets, spying, and health care.
He told CNBC, “It will raise fundamental problems about privacy, legality, ethics, and economics.”
