First, the good news. At the 15th annual Arm TechCon in San Jose, Arm announced the formation of the Autonomous Vehicle Computing Consortium (AVCC). Members include General Motors, Toyota, DENSO, Continental, Bosch, NXP, Nvidia and, of course, Arm. This cooperation among companies involved in autonomous vehicles (AV) is a good idea, considering making AVs requires input from multiple disciplines.

Look at the list: Two components suppliers, three processor core providers, two car manufacturers and one tire manufacturer. Tire manufacturer? OK we’ll let that one go for a moment. The two auto makers do not crack any kind of leader list in the field of AVs. GM might have until they announced, as did Ford, that they are delaying rollout of the AV products indefinitely, after promising to deliver in 2019. The rest take their cues from the auto manufacturers.


Recommended
Arm-led Consortium to Set APIs for AV Hardware


In GM’s defense, one of their Cadillac products cracked a few top 20 lists, but those lists were dominated by Mercedes Benz, BMW, Volvo, Tesla, Audi and VW, none of whom are in the consortium. Before the announcement GM’s Cruise division was rated just behind Waymo, who is also missing from the consortium, but will probably drop in the next ranking. Also missing is Amazon, Google, Aptix and a slough of software companies focused on developing AI.

I’m still confused about the tire company.

Further confusing the issue is the nomenclature. Dipti Vachani, senior vice president and general manager for Automotive and IoT Line of Business at Arm, in her keynote, used the term “machine learning AI” several times and the phrase was repeated by a few other Arm representatives during the day. It’s confusing because you can’t have an AI without machine learning.

An AI, to be correctly called an AI, needs three things: Data mining, machine learning, and deep learning. There are companies that do one or two of those things and they call their product and AI, but if it is missing one of those three it is two separate products and neither can be called AI. So the question is, will the consortium be working on machine learning that can be plugged into the other two from other sources, or are they making a stand alone product that isn’t actually an AI?

I tend to think the latter is the case and it makes sense because with the limited knowledge of the AVCC members about AI, they might actually think they are making one. That’s not a knock on Arm or the consortium members. It is just the state of the industry: the people actually making AI technology are not actually the people implementing it. That’s why you get the interesting demo videos of the Tesla summon future where the car freaks out over a shadow and running stop signs.

I’ve been talking to independent industry experts on the issue of AI and security in automotive systems for several weeks (a few of the interviews are on my podcast, Crucial Tech) and a major obstacle to producing autonomous cars are not just in the issues of compute challenges, which Arm and their partners seem to have that issue under control. The problem is in the AI and the reality the none of it is close to being standardized for any application, much less autonomous cars. That makes the idea of the AVCC admirable but the reality is the consortium lacks the experience and understanding of AI to make any forward progress.

Now I’m pretty sure that the members would disagree with me, especially the graphics processor partner, Nvidia. So I put the question above (what is machine learning AI?) to Arm and Nvidia, the Nvidia spokesperson went on the explain that machine learning is a subset of artificial intelligence (which I already knew and had explained in setting up my question). But he also revealed that his company is focused on the machine learning component of AI, nothing else. When I pressed the Arm spokesperson, he admitted, “Well, we are still a bit squishy on the term, AI.”

Which kinda makes my point. We are not ready for standards bodies or consortiums made up of companies on the periphery of the main problem.

I welcome all comments and brickbats. And check out the interviews at Crucial Tech.