‘AI’ is a broad umbrella term referring to any technique aimed at approximating human cognition using machines----encompasses ML, neural networks, data mining, expert systems, robotics, and more.
Chiles ’19 [John and Clara Reyes; 2019; Partner at Burr & Forman LLP; Assistant Professor of Law and Director of R&D at Michigan State Law School and Faculty Associate at Berkman Klein Center for Internet and Society at Harvard University; “Robot Loans: Artificial Intelligence and Its Place in Consumer Finance?,” 73 Consumer Fin. L.Q. Rep. 205, WestLaw]
John: Carla is an Assistant Professor of Law and Director of R&D at Michigan State Law School. She's a Faculty Associate at Berkman Klein Center for Internet and Society at Harvard Law School. Or is it at the university?
Carla: University.
*210 John: University, not the law school, which is a think tank collaboration of a number of top intellectuals in this field of fintech. Her focus is on the intersection of blockchain and law, but she knows AI, because she's authoring a casebook on AI and the law. And she's very familiar with all the writings on this issue, so she's an asset to our discussion here.
I want to do something that. ... It may seem elementary to some of you, but having sat through a number of discussions of the use of AI and machine learning and large datasets and analysis for credit-related issues, I was puzzled exactly how this works, to be honest with you. How does the AI model work? How does the machine learning model work? And what is done with big datasets, particularly non-traditional data that's being used more and more by fintech companies? As a kickoff to this, I wanted to ask Carla to give us an overview of how these particular programs and tools, that are being developed in fairly substantial quantities and numbers, how these actually work and how the machine learning and the large dataset analysis is deployed. Just how it works, so we can understand.
Carla: So, in other words, you're asking the professor to do what professors do, which is lay the foundation definitionally. I can do this. Generally speaking, there's really no straightforward consensus on the definition of artificial intelligence. As a result, discussion of AI and its sub-fields really are susceptible to a lot of hype and confusion. For instance, people use the term “machine learning” and “artificial intelligence” interchangeably, but they're really not the same thing, and so the goal here is just to break some of that down a little bit. At a pretty high level still, but just to lay the foundation definitionally.
A best consensus definition of artificial intelligence is that it's a broad term, an umbrella term, used to refer to a large set of information science. It's best understood as a set of techniques aimed at approximating some aspect of human cognition using machines. And the main takeaway is that, although we use the term “artificial intelligence” as though it's monolithic, it's really not. It covers a range of sub fields, which include, and I should say “but is not limited to” as lawyers say, neural networks, vision, data mining, expert systems, robotics, natural language processing, natural language understanding, planning, and evolutionary computation.
We actually have a class on evolutionary analysis and the law at MSU Law School, so all of these are relevant to the legal field and legal issues, but the one primarily people think about and mean when they use the term “artificial intelligence” is machine learning. Machine learning refers generally to the capacity of a system to improve its performance at some task, some specific task, over time. Often, the task involves recognizing patterns in datasets. Although machine learning outputs can include everything from translating languages to grasping objects or helping to drive a car.
*211 Specifically, machine learning relies on algorithms to analyze huge datasets. It basically performs predictive analytics far faster than any human can on a dataset that's larger than any human could reasonably get through efficiently. Notably, however and something I think that gets lost sometimes in legal discussions, is that, generally speaking, oftentimes, not always but often, the machine learning system is both ... and this should be always, is both designed and created by humans, and then you put inputs into it and it output results. Those outputs are then interpreted and applied by humans, right?