The NHTSA May Have Required Apple to Open Up Its AI Research

3 minute read
| Editorial

Autonomous car concept

Right after Apple revealed more of its plans to the U.S. Government regarding its autonomous car project, we learn that Apple is going to break with tradition and start publishing its AI research. This is an interesting sequence of events. What may have been the cause of Apple’s new approach?

What you are about to read is some basic facts followed by speculation.

The NHTSA and AI

First, the facts. As we know, most major car companies have ambitions to build autonomous, that is, self-driving cars. It will be a very competitive industry with many risks and much at stake. Safety regulations and liability issues come into play.

Also, this is a complex technology that involves considerable mastery of many kinds of senors and an artificial intelligence agent to navigate the car safely. While most companies are fairly open about their ambitions, heretofore, Apple hasn’t even admitted publicly that an autonomous car is in its product plans. This has created various problems for Apple.

Along the way, the National Highway Traffic Safety Administration (NHTSA) has become involved. Of course, while autonomous cars are expected to save many lives, the technology of autonomous cars could also fail from time to time, possibly killing or injuring passengers or others. Safety regulation is in order.

An Interesting Sequence of Events

1. Back in September, the NHTSA published its Federal Automated Vehicles Policy. For some background, here’s a good article at ars technica that explains the guidance. “The federal self-driving vehicles policy has finally been published.”

2. Last week, the Wall Street Journal revealed Apple’s ambitions. Tech Crunch provided some insight. “Apple reveals autonomous vehicle ambitions in letter to US regulators.” Notably, Tech Crunch writes:

Apple has publicly revealed its ambitions to play in the emerging market of self-driving vehicles with a policy recommendation letter to the National Highway Traffic Safety Administration.

This recommendation has the internet abuzz with the idea that Apple will make its own brand of self-driving cars, after all, not just the software or operating systems within them.

Apple’s letter discussed the philosophy of its product design and also discussed the importance of customer privacy in these two statements. Apple wrote:

“Apple uses machine learning to make its products and services smarter, more intuitive, and more personal. The company is investing heavily in the study of machine learning and automation, and is excited about the potential of automated systems in many areas, including transportation.

… and …

Apple agrees that companies should share de-identified scenario and dynamics data from crashes and near-misses…By sharing data, the industry will build a more comprehensive dataset than any one company could create alone…

[But] Data sharing should not come at the cost of privacy. Apple believes that companies should invest the resources necessary to protect individuals’ fundamental right to privacy.”

The Break

3. Next came the bombshell. Today, we learned from The Verge that Apple is going to break with tradition and start publishing its AI research.

Russ Salakhutdinov, the company’s director of artificial intelligence research reportedly made the announcement earlier today at the NIPS conference in Barcelona, Spain.

The goal, according to Salakhutdinov, is to help attract and retain top talent in the industry. Unlike some areas in tech Apple can afford to keep behind closed doors, AI thrives on open collaboration.

Russ Salakhutdinov

Russ Salakhutdinov (via Twitter)

This almost sounds like a cover story. Apple has always kept its research secret. As a result, however, Apple’s competitors are getting all the notoriety and mindshare for their AI research. For example, “Google, Facebook, and Microsoft Are Remaking Themselves Around AI.” Why would Apple change course now?

While it’s true that Apple has to compete in the open market for the best AI researchers, and it’s also true that Apple can’t recruit the best talent without letting them develop (instead of ruin) their careers, there just may be something else at play that nudged Apple over the edge.

Speculation

My suspicion is that, in Mr. Salakhutdinov’s discussions with the NHTSA, the government has expressed some concern about Apple’s development of an autonomous car in complete secret. The key to car safety is understanding and diagnosing the causes of failures (meaning lives lost or injuries thanks to software bugs). That implies an open and collegial development of AI techniques (both source code and accident analysis databases) available to the public and academic scrutiny.

Most certainly, Mr. Salakhutdinov’s earliest conviction was that Apple’s traditional insistence on secret product development would adversely affect the quality of the people he could hire. But my surmise is that he couldn’t close that case with CEO Tim Cook until the NHTSA expressed the view that cars with AI agents designed in complete secret couldn’t be approved for public use. Apple’s research, in this technology, would have to be more open to scrutiny.

And with that, Apple gets a win-win. It makes perfect sense to me.

2 Comments Add a comment

  1. wab95

    John:

    I concur with you and geoduck, not only is this plausible, but probable. In the world of licensing and certification by regulators, documentation rules. Indeed, I’ve been wondering if this would not ultimately have to happen.

    In clinical trials, for example, not only does one have to follow GCP (Good Clinical Practice) procedure, one has to document the training, the standardisation exercises for consistency of results, every step of approval for the protocol and every key step of the protocol, including the consenting of participants, and have it checked and verified by independent clinical trial monitors including the data management, and provide as much supportive evidence for your results as possible, including related peer-reviewed publications. Even one missing or questionable piece of that evidence can jeopardise the ultimate licensing of the intervention under study.

    It stands to reason, then, that Apple are being compelled to document in the public domain key elements of their product and the key steps to its ultimate licensing or certification, as well as to attract the type of talent necessary to make it so.

    In the long run, this is better not only for Apple, but the industry, the public and the consumer.

Add a Comment

Log in to comment (TMO, Twitter, Facebook) or Register for a TMO Account