Artificial Intelligence Swarms Silicon Valley on Wings and Wheels | Hacker News Reader

Artificial Intelligence Swarms Silicon Valley on Wings and Wheels

nytimes.com

37 pointshvo12 comments

That inventory-checking robot may be the most useful AI startup product I've seen. It's a valuable and large problem, challenging yet tractable with today's technology.

Curious if anyone knows other valuable and unique markets being addressed by new startups.

The inventory-checking robot is an interesting pivot for Bossa Nova, a company that is 11 years old and originally focused on robots that could balance on a single ball (it spun out of Carnegie Mellon where the original Ballbot was created). They struggled to monetize it and made toys, which they then sold off and tried to figure out what to do. They ultimately moved away from their special technology entirely to work on inventory. It seems like a huge market, although I wonder how well it works since its vision is largely limited to the shelf item at the front and it likely cannot tell how many there are most of the time. A much simpler solution I've seen at many stores is to have a simple electronic system for each item and it keeps count of how many there are, I imagine each of these costs around a dollar (although you need hundreds or thousands). It's also simple and reliable, of course the robot scales easier for mega-stores assuming it can handle larger inventories easily.

I think Mashgin has a really fascinating idea which is to automate the checkout counter at grocery stores, although I know nothing about how their algorithms work.

I'm thinking, isn't the robot somewhat unnecessary? Can't it be done by software instead? Maybe connecting into the inventory management system / POS if the grocery store and do the AI processing in it.

I thought that, but the inventory management gets out of sync when stock is removed without going through the POS (stealing)

Even for a warehouse (restricted access), there is a need for stock counting as mistakes can be made

Eventually, an AI that accounted for shrink would be more accurate than any number of cycle counts. Actually, it would not take long, most major retailers have shrink predictions down to two or three decimals, you would just need to train the AI to account for shrink. The most important role of inventory is knowing how much to re-order, which can be done by the AI more easily than any human analysts once you have enough training data. Source: worked in retail management, then inventory software.

For many readers of HN, this article is obviously true. But so obvious that you might wonder if it's newsworthy, a rehashing of what we hear so often:

* AI will take jobs

> a prototype inventory checker developed by Bossa Nova Robotics silently glided through the aisles using computer vision to automatically perform a task that humans have done manually for centuries.

* VCs move in packs

> “We saw a slow trickle in investments in robotics, and suddenly, boom — there seem to be a dozen companies ..."

* The rush for talent

> “The number of people trying to get the students to drop out of the class halfway through because now they know a little bit of this stuff is crazy.”

* The strawman of superintelligence

> For some technologists, today’s technical advances are laying the groundwork for truly brilliant machines that will soon have human-level intelligence.

* The evocation of AI winter, just to be safe

> ...so the commercial enthusiasm ended in disappointment, leading to a period now referred to as the “A.I. Winter.”

Add a potted history of Silicon Valley as a "place that reinvents itself" and the Canadian mafia's origin story about deep learning, and you have a recipe for almost any news article about AI, ever.

Not much new here.

Take that framework, put it into a program, and pretty soon you could have an AI that churns out articles for the NYT!

Are drones and robots really about AI or are they more about processors that can handle intensive image processing and control systems on a cheap, low-power silicon die?

Don't get me wrong, I think robotics is gonna be yuuuuuge, but I see AI as more of a software play in the near term (consumer/biz, not embedded.)

Oh, and robots don't have a minimum wage. Something to think about as we push up labor costs at the low end (unnecessarily IMO.) Is it a coincidence that I see touch-screens in SF McDonalds the year following the min-wage increase? The technology has been there for a long time.

The tricky part is understanding consumer behavior. McDonald's has been rolling out touchscreens in San Francisco where they are still operating, but they also tried this in Colorado and in under a month they removed most because they were losing customers. Operational efficiencies are always wonderful on the backend, but when you put them in front of your customers it's important to understand their reactions as they may have negative reactions. There are lots of people who are not happy about robots taking jobs, and if they can see these robots it will influence their loyalty to that store.

Good point. I did walk out of a McDonalds w/ touch screens by Union Square a few months ago. It wasn't a statement on robots. I just didn't find the experience very good. I remember being a bit confused by the process.

Funny, but now I remember ordering at touch screens at McDonalds in France last year. Even though it was in French (I speak no French) I found the process simple. Simpler than face to face ordering, to be sure, since I could order by picture.

Maybe, like the self-checkout lines at the grocery store, which I hated at first but usually choose now because the line is quicker, there's a learning curve. In fact, maybe McDonalds in Colorado should have waited before pulling those touchscreens?

I there is more success than 1980s A.I. A whole lot hoopla then. Fear Japan would whomp us. And then never went anywhere. In the 1980s it was LISP machines, expert systems, and logical programming languages. Now it is deep learning and self driving cars.

The inventory robot reminds me of an idea I had 11 years ago to use roomba tech + rfid readers to track inventory in stores like walmart. Curious to see how their use of computer vision can account for things like depth of shelves etc.

Reply on news.ycombinator.com