Deep learning perceptions are not based on many facts, but on common predictions other past technologies faced
BY:
Cassandra Este
Social Media Analyst
PROJECT COUNSEL MEDIA
20 September 2022 (San Francisco, California) – As technology advances, people speculate about how it will change society, especially in the twentieth century. We were supposed to have flying cars, holograms would be a daily occurrence, and automation would make most jobs obsolete.
Yet here we are in the twenty-first century and futurists only got some of the predictions right. It begs the question if technology developers, such as deep learning researchers, are overhyping their industry. AI Snake Oil (what a great name) explores the idea in an article titled “Why Are Deep Learning Technologists So Overconfident?”
According to the authors Arvind Narayanan and Says Kapoor, the hype surrounding deep learning is similar to past and present scientific dogma: “a core belief that binds the group together and gives it its identity.” Deep learning researchers’ dogma is that learning problems can be resolved by collecting training examples. It sounds great in theory, but simply collecting training examples is not a complete answer.
It does not take much investigation to discover that deep learning training datasets are rich in biased and incomplete information. Deep learning algorithms are incapable of understanding perception, judgment, and social problems. Researchers describe the algorithms as great prediction tools, but it is the furthest thing from the truth.
Deep learning researchers are aware of the faults in the technology and are stuck in the same “us vs. them” mentality that inventors have found themselves in for centuries. Deep learning perceptions are not based on many facts, but on common predictions other past technologies faced:
“This contempt is also mixed with an ignorance of what domain experts actually do. Technologists proclaiming that AI will make various professions obsolete is like if the inventor of the typewriter had proclaimed that it will make writers and journalists obsolete, failing to recognize that professional expertise is more than the externally visible activity. Of course, jobs and tasks have been successfully automated throughout history, but someone who doesn’t work in a profession and doesn’t understand its nuances is in a poor position to make predictions about how automation will impact it.”
Deep learning will be the basis for future technology, but it has a long way to go before it is perfected. All advancements go through trial and error. Deep learning researchers need to admit their mistakes, invest funding with better datasets, and experiment. Practice makes perfect. When smart software goes off the rails, there are PR firms to make everything better again.
Obviously, there are self-serving reasons for any field to hype itself. But that doesn’t explain all of it, and many deep learning people genuinely believe their overconfident predictions. In their article, Arvind Narayanan and Says Kapoor show the cultural and historical reasons for this and note:
“With that understanding of those reasons it will help you resist the hype and push back the next time you meet a true believer of deep learning — while still acknowledging that it works well in a limited set of domains and tasks”.
And as I have noted before, there is a tendency across different subfields in AI to see value in a small collection of influential “benchmarks”, which many will term “general benchmarks”. But you need to understand how these benchmarks operate merely as stand-ins or abstractions for a range of anointed common problems that are frequently framed as “foundational milestones” on the path towards flexible and generalizable AI systems. You really need to study and understand the calcified misinformation and bias and bad data that abounds, and understand how these “benchmarks” are designed, constructed and used in order to see their framing – or you will never “get” the deep limitations of AI or deep learning.
