The most popular and comprehensive Open Source ECM platform
Let us know more
Artificial Intelligence: Foundational Models Come under Fire
Foundational Models in artificial intelligence are generalized models that have been pre-trained on huge amounts of data and have millions or even billions of parameters that can be tuned. The models then are adopted and tuned for solving specific problems.
A recent example of a foundational model is OpenAI’s GPT-3. GPT-3 is able to respond to prompts. It has been configured and tuned to be able to translate language and even to write software code.
The foundational models are driving much of the current developments in AI. In many ways the results that AI algorithms have been able to produce have been amazing in accuracy and sophistication. But there is a worry about using these algorithms. Exactly how these algorithms work is still not fully understood and many people point to the fact that the data used to train the models is often biased and that the models only further propagate the biases. Other questions that arise relate to safety, fairness and reliability.
Recently, a draft proposal on foundational models was published by Stanford Human-Centered Artificial Intelligence’s (HAI) Center for Research on Foundation Models (CRFM). The proposal attempts to draw together an interdisciplinary team of students, faculty and researchers in the Stanford community to look at the legal, ethical, and economic questions that AI foundational models raise.
But some object just to the use of the term ‘foundational model’ and find problematic the idea that substantial investment should be made to build on top of something that still not mature or well understood.
Jitendra Malik, a professor at UC Berkeley, said that “I think the term ‘foundation’ is horribly wrong. These models are really castles in the air; they have no foundation whatsoever. The language we have in these models is not grounded, there is this fakeness, there is no real understanding.”
