Trillions of parameters

Maria Deutscher, writing for Silicon Angle:

Microsoft Corp. has released a new version of its open-source DeepSpeed tool that it says will enable the creation of deep learning models with a trillion parameters, more than five times as many as in the world’s current largest model.

Microsoft AI tool enables ‘extremely large’ models with a trillion parameters

That’s a lot of transformations. If there’s a pattern, a trillion parameters should be able to find and store it.

Portland bans facial recognition by private entities

34.10.030 Prohibition.

Except as provided in the Exceptions section below, a Private Entity shall not use Face Recognition Technologies in Places of Public Accommodation within the boundaries of the City of Portland.

34.10.040 Exceptions.

The prohibition in this Chapter does not apply to use of Face Recognition Technologies:

1. To the extent necessary for a Private Entity to comply with federal, state, or local laws;

2. For user verification purposes by an individual to access the individual’s own personal or employer issued communication and electronic devices; or

3. In automatic face detection services in social media applications.

Prohibit the use of Face Recognition Technologies by Private Entities in Places of Public Accommodation in the City (via PRIVACY & INFORMATION SECURITY LAW BLOG)

Note the exception for use in “social media applications.”

What does it mean for AI to be “explainable”?

A NIST paper attempts to answer this question:

Briefly, our four principles of explainable AI are:

Explanation: Systems deliver accompanying evidence or reason(s) for all outputs. 

Meaningful: Systems provide explanations that are understandable to individual users. 

Explanation Accuracy: The explanation correctly reflects the system’s process for generating the output. 

Knowledge Limits: The system only operates under conditions for which it was designed or when the system reaches a sufficient confidence in its output. 

Four Principles of Explainable Artificial Intelligence

Stating this differently: there should be an explanation, it should be understandable and accurate, and the system should stop when it’s generating nonsense.

These are very reasonable principles, but likely tough to deliver with current technology.

Indeed, the paper discusses that humans are often unable to explain why they have taken a certain action:

People fabricate reasons for their decisions, even those thought to be immutable, such as personally held opinions [24, 34, 99]. In fact, people’s conscious reasoning that is able to be verbalized does not seem to always occur before the expressed decision. Instead, evidence suggests that people make their decision and then apply reasons for those decisions after the fact [95]. From a neuroscience perspective, neural markers of a decision can occur up to 10 seconds before a person’s conscious awareness [85]. This finding suggests that decision making processes begin long before our conscious awareness. 

Id. at 14.

And it is well documented that even experts generally cannot predict their own accuracy.

What hope do the AI’s have?

AlphaDogfight wins 5-0 in F-16 battle vs human

Will Knight, writing for Wired:

Last week, a technique popularized by DeepMind was adapted to control an autonomous F-16 fighter plane in a Pentagon-funded contest to show off the capabilities of AI systems. In the final stage of the event, a similar algorithm went head-to-head with a real F-16 pilot using a VR headset and simulator controls. The AI pilot won, 5-0.

A Dogfight Renews Concerns About AI’s Lethal Potential

This is an under-discussed issue, but is inevitable. DeepMind is convinced that its AlphaZero DNN can master any two-player, turn-based game that shows perfect information. And its AlphaStar DNN shows what it can do in real-time games as well. It is a natural, and inevitable, extension to war capabilities.

Is this ok? Does that question even matter? How long before human-in-the-loop is the unacceptable bottleneck?

Freedom vs. Security continued…

Kashmir Hill for the NYT:

Floyd Abrams, one of the most prominent First Amendment lawyers in the country, has a new client: the facial recognition company Clearview AI.

Litigation against the start-up “has the potential of leading to a major decision about the interrelationship between privacy claims and First Amendment defenses in the 21st century,” Mr. Abrams said in a phone interview. He said the underlying legal questions could one day reach the Supreme Court.

Facial Recognition Start-Up Mounts a First Amendment Defense

Is everything known to the public truly available for any use whatsoever? We are trending away from that view, and this will be a battle to watch closely.

Facial recognition software countermeasures

Software that tweaks photos to hide them from facial recognition:

A start-up called Clearview AI, for example, scraped billions of online photos to build a tool for the police that could lead them from a face to a Facebook account, revealing a person’s identity.

Now researchers are trying to foil those systems. A team of computer engineers at the University of Chicago has developed a tool that disguises photos with pixel-level changes that confuse facial recognition systems.

This Tool Could Protect Your Photos From Facial Recognition

This is of course just an arms race. The facial recognition will improve, the hiding software will get tweaked.

Rite Aid has been using facial recognition for 8 years

Jeffrey Dastin writing for Reuters:

The cameras matched facial images of customers entering a store to those of people Rite Aid previously observed engaging in potential criminal activity, causing an alert to be sent to security agents’ smartphones. Agents then reviewed the match for accuracy and could tell the customer to leave.

Rite Aid deployed facial recognition systems in hundreds of U.S. stores

The DeepCam systems were primarily deployed in “lower-income, non-white neighborhoods,” and, according to current and former Rite Aid employees, a previous system called FaceFirst regularly made mistakes:

“It doesn’t pick up Black people well,” one loss prevention staffer said last year while using FaceFirst at a Rite Aid in an African-American neighborhood of Detroit. “If your eyes are the same way, or if you’re wearing your headband like another person is wearing a headband, you’re going to get a hit.”

Shortcuts in Learing

So starts a great article on DNN’s and learning shortcuts:

Recently, researchers trained a deep neural network to classify breast cancer, achieving a performance of 85%. When used in combination with three other neural network models, the resulting ensemble method reached an outstanding 99% classification accuracy, rivaling expert radiologists with years of training.

The result described above is true, with one little twist: instead of using state-of-the-art artificial deep neural networks, researchers trained “natural” neural networks – more precisely, a flock of four pigeons – to diagnose breast cancer.

Shortcuts: How Neural Networks Love to Cheat

Sometimes the learning is clever; sometimes the learning is a problem. Mostly though, it’s hard to make sure an AI learns what you want it to learn.