• John Ennis

Eye on AI - September 19th, 2019

Welcome to Aigora's "Eye on AI" series, where we round up exciting news at the intersection of consumer science and artificial intelligence!

This week, we look at two important articles that delve into two major challenges AI systems are encountering as they integrate with the modern economy.

The Machine-Learning Reproducibility Problem

In the first article, titled “Artificial Intelligence Confronts a 'Reproducibility' Crisis”, author Gregory Barber describes how many machine-learning systems have become like black boxes in that they work, but are nearly impossible to understand.

“It’s one thing to marvel at the eloquence of a new text generator or the “superhuman” agility of a videogame-playing bot,” writes Barber. “But even the most sophisticated researchers have little sense of how they work. Replicating those AI models is important not just for identifying new avenues of research, but also as a way to investigate algorithms as they augment, and in some cases supplant, human decision-making....”

Some researchers have tried to implement processes to ensure bread crumbs are left behind once winning models are found so they can understand how the high-performing model work, but these processes often fall short. Much of the information needed for reproduction is often omitted in reports for proprietary reasons, and researchers are hesitant to impose hard new standards as they might lead to research constraints.

Joelle Pineau, a computer science professor at McGill, remains optimistic a solution can be found given the young industry’s flexibility. Only time will tell.

The One-Off AI Problem

In the second article, titled “Artificial Intelligence Only Goes So Far In Today’s Economy, Says MIT Study”, author Joe McKendrick describes how a new MIT study reveals that AI’s evolution is stunted by the unreliability of data.

The problem, notes McKendrick, is the data – both the amount needed, and the ways in which it’s used within new machine-learning systems.

“The industries that use ML are slowly learning that the data used to train ML systems must be as unbiased and trusted as the systems themselves need to be—crucial challenges in an era of hacking and cyber-warfare,” writes the co-authors of the featured MIT report in the article, titled Work of the Future. “..... We are a long way from AI systems that can read the news, re-plan supply chains in response to anticipated events like Brexit or trade disputes, and adapt production tasks to new sources of parts and materials.”

Explainability is a central issue, – see ‘black box’ article above – all of this adding up to AI having a long way to go before it can become fully automated. For those expecting the world to be run by machines within the next few years, you may be waiting longer than you expect.

Other news:

That's it for now. If you'd like to receive email updates from Aigora, including weekly video recaps of our blog activity, click on the button below to join our email list. Thanks for stopping by!

©2020 Aigora