Hi everyone,
It’s Masters Week which means it’s going to be a great week. Nothing can rain on your parade and get you down … unless, you’re Amazon’s head of innovation – and people were to learn that you secretly had over a 1,000 outsourced people doing the work your AI was allegedly doing.
We’ve talked a lot about the upside of AI in recent weeks. With the Amazon Just Walk Out news, we thought it would be a good time to share an opposing view on the all the hope and promise. Over the past decade, this wouldn’t be the first time we’ve heard a lot of hype about AI disruption – only for the reality to fall short.
That was the headline of Ars Technica’s article covering the story of a damning report on what’s actually going on at Amazon’s Just Walk Out.
The Just Walk Out program was a brainchild of Jeff Bezos who wanted to eliminate the worst part about retail shopping – the checkout. Translation: AI will replace cashiers.
So, by installing hundreds of expensive cameras and professional lighting into a small 1,000-2,000 square foot store, Amazon’s AI could automatically identify what a shopper picks up from the shelf. The shopper can walk out of the store and never have to go to a cashier. They can “just walk out.”
Unfortunately, it turns out that after hundreds of stores and millions of dollars of investment the program never actually worked.
“Amazon had more than 1,000 people in India working on Just Walk Out as of mid-2022 whose jobs included manually reviewing transactions and labeling images from videos to train Just Walk Out’s machine learning model.” Recently published report by The Information captures the inside scoop on what was really happening at Amazon’s Just Walk Out program.
Remember Self Driving Cars?
Back in 2017 and 2018 when we were working with Ford, every other week there was news about an autonomous technology startup raising money, or getting acquired, or a new press release about Level 5 autonomy coming “in the next few years.” Just like the cashiers in Just Walk Out, AI was going to fully replace human drivers.
Automakers spent billions of dollars. Google, Baidu, Intel and other tech companies spent even more than that.
With all this investment pouring in, a top management consulting firm predicted that we would have fully autonomous (i.e., level 5) autonomy commercially introduced by 2025.
80 or 90% accurate just isn’t enough.
Has AI had an impact in retail, automotive, and medicine? Absolutely. And we expect that impact to only grow in years ahead. But these past hype cycles are object lessons in why its healthy to have some skepticism about the promises of the current AI hype cycle. As we’ve written about in recent weeks, AI already is making a big impact across B2B distribution. But don’t expect an “autonomous” distributor anytime soon.
For critical functions, like driving, medicine and distribution, a lack of accuracy means AI cannot be truly autonomous yet. Getting to the right level of accuracy for these critical functions is really, really hard – and it’s not clear that the current generations of AI models can truly get there without further tech breakthroughs or different approaches. AI + humans will be more efficient than humans alone, but AI isn’t replacing the human in many critical scenarios any time soon.
If the car can drive itself 95% of the time really well, that 5% “whoops” is too large of a margin of error. For Amazon’s Just Walk Out, they weren’t able to get to even 80% accuracy. Hence the need for thousands of off-shore workers to fill the gaps.
In order for AI to actually deliver on the hope and promise, it needs to be materially more accurate than the human alternative.
Are hallucinations the next Just Walk Out?
Keep reading with a 7-day free trial
Subscribe to B2B Distribution Technology to keep reading this post and get 7 days of free access to the full post archives.