How AI struggles with bike lanes and bias

2 years ago 253

Commentary: Despite continued advances successful AI, we inactive haven't solved immoderate of its astir basal problems.

artificial-intelligence-1.jpg

Image: iStock/Jolygon

We've been truthful disquieted astir whether AI-driven robots volition instrumentality our jobs that we forgot to inquire a overmuch much basal question: volition they instrumentality our motorcycle lanes?

That's the question Austin, Texas, is presently grappling with, and it points to each sorts of unresolved issues related to AI and robots. The biggest of those? As revealed successful Anaconda's State of Data Science 2021 report, the biggest interest data scientists person with AI contiguous is the possibility, adjacent likelihood, of bias successful the algorithms. 

SEE: Artificial quality morals policy (TechRepublic Premium)

Move over, robot

Leave it to Austin (tagline: "Keep Austin weird") to beryllium the archetypal to person to grapple with robot overlords taking implicit their motorcycle lanes. If a robot that looks similar a "futuristic crystal pick truck" successful your lane seems innocuous, see what Jake Boone, vice-chair of Austin's Bicycle Advisory Council, has to say: "What if successful 2 years we person respective 100 of these connected the road?"

If this seems unlikely, see conscionable however accelerated electrical scooters took implicit galore cities.

The problem, then, isn't truly 1 of a radical of Luddite bicyclists trying to hammer distant progress. Many of them admit that 1 much robotic transportation conveyance is 1 little car connected the road. The robots, successful different words, committedness to alleviate postulation and amended aerial quality. Even so, specified benefits person to beryllium weighed against the negatives, including clogged motorcycle lanes successful a metropolis wherever infrastructure is already stretched. (If you've not been successful Austin postulation recently, well, it's not pleasant.)

As a society, we haven't had to grapple with issues similar this. Not yet. But if "weird" Austin is immoderate indicator, we're astir to person to deliberation cautiously astir however we privation to clasp AI and robots. And we're already precocious successful coming to grips with a overmuch bigger contented than motorcycle lanes: bias.

Making algorithms just

People conflict with bias, truthful it's not astonishing the algorithms we constitute do, excessively (a occupation that has persisted for years). In fact, inquire 3,104 information scientists (as Anaconda did) to sanction the biggest occupation successful AI today, and they'll archer you it's bias (Figure A).

Figure A

ai-problems.jpg

Image: Anaconda

That bias creeps into the information we take to cod (and keep), arsenic good arsenic the models we deploy. Fortunately, we admit the problem. Now what are we doing astir it?

Today, conscionable 10% of survey respondents said their organizations person already implemented a solution to amended fairness and bounds bias. Still, it's a affirmative motion that 30% program to bash truthful wrong the adjacent 12 months, compared to conscionable 23% successful 2020. At the aforesaid time, 31% of respondents said they don't presently person plans to guarantee exemplary explainability and interpretability (which permeability would assistance to mitigate against bias), 41% said they've already started to enactment connected doing so, oregon program to bash truthful wrong the adjacent 12 months. 

So, are we determination yet? No. We inactive person tons of enactment to bash connected bias successful AI, conscionable arsenic we request to fig retired much pedestrian topics similar postulation successful motorcycle lanes (or fault successful car accidents involving self-driving cars). The bully news? As an industry, we're alert of the occupation and progressively moving to hole it. 

Disclosure: I enactment for AWS, but the views expressed herein are mine.

Data, Analytics and AI Newsletter

Learn the latest quality and champion practices astir information science, large information analytics, and artificial intelligence. Delivered Mondays

Sign up today

Also spot

Read Entire Article