Tag Archives: intelligence

#436202 Trump CTO Addresses AI, Facial ...

Michael Kratsios, the Chief Technology Officer of the United States, took the stage at Stanford University last week to field questions from Stanford’s Eileen Donahoe and attendees at the 2019 Fall Conference of the Institute for Human-Centered Artificial Intelligence (HAI).

Kratsios, the fourth to hold the U.S. CTO position since its creation by President Barack Obama in 2009, was confirmed in August as President Donald Trump’s first CTO. Before joining the Trump administration, he was chief of staff at investment firm Thiel Capital and chief financial officer of hedge fund Clarium Capital. Donahoe is Executive Director of Stanford’s Global Digital Policy Incubator and served as the first U.S. Ambassador to the United Nations Human Rights Council during the Obama Administration.

The conversation jumped around, hitting on both accomplishments and controversies. Kratsios touted the administration’s success in fixing policy around the use of drones, its memorandum on STEM education, and an increase in funding for basic research in AI—though the magnitude of that increase wasn’t specified. He pointed out that the Trump administration’s AI policy has been a continuation of the policies of the Obama administration, and will continue to build on that foundation. As proof of this, he pointed to Trump’s signing of the American AI Initiative earlier this year. That executive order, Kratsios said, was intended to bring various government agencies together to coordinate their AI efforts and to push the idea that AI is a tool for the American worker. The AI Initiative, he noted, also took into consideration that AI will cause job displacement, and asked private companies to pledge to retrain workers.

The administration, he said, is also looking to remove barriers to AI innovation. In service of that goal, the government will, in the next month or so, release a regulatory guidance memo instructing government agencies about “how they should think about AI technologies,” said Kratsios.

U.S. vs China in AI

A few of the exchanges between Kratsios and Donahoe hit on current hot topics, starting with the tension between the U.S. and China.

Donahoe:

“You talk a lot about unique U.S. ecosystem. In which aspect of AI is the U.S. dominant, and where is China challenging us in dominance?

Kratsios:

“They are challenging us on machine vision. They have more data to work with, given that they have surveillance data.”

Donahoe:

“To what extent would you say the quantity of data collected and available will be a determining factor in AI dominance?”

Kratsios:

“It makes a big difference in the short term. But we do research on how we get over these data humps. There is a future where you don’t need as much data, a lot of federal grants are going to [research in] how you can train models using less data.”

Donahoe turned the conversation to a different tension—that between innovation and values.

Donahoe:

“A lot of conversation yesterday was about the tension between innovation and values, and how do you hold those things together and lead in both realms.”

Kratsios:

“We recognized that the U.S. hadn’t signed on to principles around developing AI. In May, we signed [the Organization for Economic Cooperation and Development Principles on Artificial Intelligence], coming together with other Western democracies to say that these are values that we hold dear.

[Meanwhile,] we have adversaries around the world using AI to surveil people, to suppress human rights. That is why American leadership is so critical: We want to come out with the next great product. And we want our values to underpin the use cases.”

A member of the audience pushed further:

“Maintaining U.S. leadership in AI might have costs in terms of individuals and society. What costs should individuals and society bear to maintain leadership?”

Kratsios:

“I don’t view the world that way. Our companies big and small do not hesitate to talk about the values that underpin their technology. [That is] markedly different from the way our adversaries think. The alternatives are so dire [that we] need to push efforts to bake the values that we hold dear into this technology.”

Facial recognition

And then the conversation turned to the use of AI for facial recognition, an application which (at least for police and other government agencies) was recently banned in San Francisco.

Donahoe:

“Some private sector companies have called for government regulation of facial recognition, and there already are some instances of local governments regulating it. Do you expect federal regulation of facial recognition anytime soon? If not, what ought the parameters be?”

Kratsios:

“A patchwork of regulation of technology is not beneficial for the country. We want to avoid that. Facial recognition has important roles—for example, finding lost or displaced children. There are use cases, but they need to be underpinned by values.”

A member of the audience followed up on that topic, referring to some data presented earlier at the HAI conference on bias in AI:

“Frequently the example of finding missing children is given as the example of why we should not restrict use of facial recognition. But we saw Joy Buolamwini’s presentation on bias in data. I would like to hear your thoughts about how government thinks we should use facial recognition, knowing about this bias.”

Kratsios:

“Fairness, accountability, and robustness are things we want to bake into any technology—not just facial recognition—as we build rules governing use cases.”

Immigration and innovation

A member of the audience brought up the issue of immigration:

“One major pillar of innovation is immigration, does your office advocate for it?”

Kratsios:

“Our office pushes for best and brightest people from around the world to come to work here and study here. There are a few efforts we have made to move towards a more merit-based immigration system, without congressional action. [For example, in] the H1-B visa system, you go through two lotteries. We switched the order of them in order to get more people with advanced degrees through.”

The government’s tech infrastructure

Donahoe brought the conversation around to the tech infrastructure of the government itself:

“We talk about the shiny object, AI, but the 80 percent is the unsexy stuff, at federal and state levels. We don’t have a modern digital infrastructure to enable all the services—like a research cloud. How do we create this digital infrastructure?”

Kratsios:

“I couldn’t agree more; the least partisan issue in Washington is about modernizing IT infrastructure. We spend like $85 billion a year on IT at the federal level, we can certainly do a better job of using those dollars.” Continue reading

Posted in Human Robots

#436200 AI and the Future of Work: The Economic ...

This week at MIT, academics and industry officials compared notes, studies, and predictions about AI and the future of work. During the discussions, an insurance company executive shared details about one AI program that rolled out at his firm earlier this year. A chatbot the company introduced, the executive said, now handles 150,000 calls per month.

Later in the day, a panelist—David Fanning, founder of PBS’s Frontline—remarked that this statistic is emblematic of broader fears he saw when reporting a new Frontline documentary about AI. “People are scared,” Fanning said of the public’s AI anxiety.

Fanning was part of a daylong symposium about AI’s economic consequences—good, bad, and otherwise—convened by MIT’s Task Force on the Work of the Future.

“Dig into every industry, and you’ll find AI changing the nature of work,” said Daniela Rus, director of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). She cited recent McKinsey research that found 45 percent of the work people are paid to do today can be automated with currently available technologies. Those activities, McKinsey found, represent some US $2 trillion in wages.

However, the threat of automation—whether by AI or other technologies—isn’t as new as technologists on America’s coasts seem to believe, said panelist Fred Goff, CEO of Jobcase, Inc.

“If you live in Detroit or Toledo, where I come from, technology has been displacing jobs for the last half-century,” Goff said. “I don’t think that most people in this country have the increased anxiety that the coasts do, because they’ve been living this.”

Goff added that the challenge AI poses for the workforce is not, as he put it, “getting coal miners to code.” Rather, he said, as AI automates some jobs, it will also open opportunities for “reskilling” that may have nothing to do with AI or automation. He touted trade schools—teaching skills like welding, plumbing, and electrical work—and certification programs for sales industry software packages like Salesforce.

On the other hand, a documentarian who reported another recent program on AI—Krishna Andavolu, senior correspondent for Vice Media—said “reskilling” may not be an easy answer.

“People in rooms like this … don’t realize that a lot of people don’t want to work that much,” Andavolu said. “They’re not driven by passion for their career, they’re driven by passion for life. We’re telling a lot of these workers that they need to reskill. But to a lot of people that sounds like, ‘I’ve got to work twice as hard for what I have now.’ That sounds scary. We underestimate that at our peril.”

Part of the problem with “reskilling,” Andavolu said, is that some high-growth industries involve caregiving for seniors and in medical facilities—roles which are traditionally considered “feminized” careers. Destigmatizing these jobs, and increasing the pay to match the salaries of displaced jobs like long-haul truck drivers, is another challenge.

Daron Acemoglu, MIT Institute Professor of Economics, faulted the comparably slim funding of academic research into AI.

“There is nothing preordained about the progress of technology,” he said. Computers, the Internet, antibiotics, and sensors all grew out of government and academic research programs. What he called the “blue-sky thinking” of non-corporate AI research can also develop applications that are not purely focused on maximizing profits.

American companies, Acemoglu said, get tax breaks for capital R&D—but not for developing new technologies for their employees. “We turn around and [tell companies], ‘Use your technologies to empower workers,’” he said. “But why should they do that? Hiring workers is expensive in many ways. And we’re subsidizing capital.”

Said Sarita Gupta, director of the Ford Foundation’s Future of Work(ers) Program, “Low and middle income workers have for over 30 years been experiencing stagnant and declining pay, shrinking benefits, and less power on the job. Now technology is brilliant at enabling scale. But the question we sit with is—how do we make sure that we’re not scaling these longstanding problems?”

Andrew McAfee, co-director of MIT’s Initiative on the Digital Economy, said AI may not reduce the number of jobs available in the workplace today. But the quality of those jobs is another story. He cited the Dutch economist Jan Tinbergen who decades ago said that “Inequality is a race between technology and education.”

McAfee said, ultimately, the time to solve the economic problems AI poses for workers in the United States is when the U.S. economy is doing well—like right now.

“We do have the wind at our backs,” said Elisabeth Reynolds, executive director of MIT’s Task Force on the Work of the Future.

“We have some breathing room right now,” McAfee agreed. “Economic growth has been pretty good. Unemployment is pretty low. Interest rates are very, very low. We might not have that war chest in the future.” Continue reading

Posted in Human Robots

#436188 The Blogger Behind “AI ...

Sure, artificial intelligence is transforming the world’s societies and economies—but can an AI come up with plausible ideas for a Halloween costume?

Janelle Shane has been asking such probing questions since she started her AI Weirdness blog in 2016. She specializes in training neural networks (which underpin most of today’s machine learning techniques) on quirky data sets such as compilations of knitting instructions, ice cream flavors, and names of paint colors. Then she asks the neural net to generate its own contributions to these categories—and hilarity ensues. AI is not likely to disrupt the paint industry with names like “Ronching Blue,” “Dorkwood,” and “Turdly.”

Shane’s antics have a serious purpose. She aims to illustrate the serious limitations of today’s AI, and to counteract the prevailing narrative that describes AI as well on its way to superintelligence and complete human domination. “The danger of AI is not that it’s too smart,” Shane writes in her new book, “but that it’s not smart enough.”

The book, which came out on Tuesday, is called You Look Like a Thing and I Love You. It takes its odd title from a list of AI-generated pick-up lines, all of which would at least get a person’s attention if shouted, preferably by a robot, in a crowded bar. Shane’s book is shot through with her trademark absurdist humor, but it also contains real explanations of machine learning concepts and techniques. It’s a painless way to take AI 101.

She spoke with IEEE Spectrum about the perils of placing too much trust in AI systems, the strange AI phenomenon of “giraffing,” and her next potential Halloween costume.

Janelle Shane on . . .

The un-delicious origin of her blog
“The narrower the problem, the smarter the AI will seem”
Why overestimating AI is dangerous
Giraffing!
Machine and human creativity

The un-delicious origin of her blog IEEE Spectrum: You studied electrical engineering as an undergrad, then got a master’s degree in physics. How did that lead to you becoming the comedian of AI?
Janelle Shane: I’ve been interested in machine learning since freshman year of college. During orientation at Michigan State, a professor who worked on evolutionary algorithms gave a talk about his work. It was full of the most interesting anecdotes–some of which I’ve used in my book. He told an anecdote about people setting up a machine learning algorithm to do lens design, and the algorithm did end up designing an optical system that works… except one of the lenses was 50 feet thick, because they didn’t specify that it couldn’t do that.
I started working in his lab on optics, doing ultra-short laser pulse work. I ended up doing a lot more optics than machine learning, but I always found it interesting. One day I came across a list of recipes that someone had generated using a neural net, and I thought it was hilarious and remembered why I thought machine learning was so cool. That was in 2016, ages ago in machine learning land.
Spectrum: So you decided to “establish weirdness as your goal” for your blog. What was the first weird experiment that you blogged about?
Shane: It was generating cookbook recipes. The neural net came up with ingredients like: “Take ¼ pounds of bones or fresh bread.” That recipe started out: “Brown the salmon in oil, add creamed meat to the mixture.” It was making mistakes that showed the thing had no memory at all.
Spectrum: You say in the book that you can learn a lot about AI by giving it a task and watching it flail. What do you learn?
Shane: One thing you learn is how much it relies on surface appearances rather than deep understanding. With the recipes, for example: It got the structure of title, category, ingredients, instructions, yield at the end. But when you look more closely, it has instructions like “Fold the water and roll it into cubes.” So clearly this thing does not understand water, let alone the other things. It’s recognizing certain phrases that tend to occur, but it doesn’t have a concept that these recipes are describing something real. You start to realize how very narrow the algorithms in this world are. They only know exactly what we tell them in our data set.
BACK TO TOP↑ “The narrower the problem, the smarter the AI will seem” Spectrum: That makes me think of DeepMind’s AlphaGo, which was universally hailed as a triumph for AI. It can play the game of Go better than any human, but it doesn’t know what Go is. It doesn’t know that it’s playing a game.
Shane: It doesn’t know what a human is, or if it’s playing against a human or another program. That’s also a nice illustration of how well these algorithms do when they have a really narrow and well-defined problem.
The narrower the problem, the smarter the AI will seem. If it’s not just doing something repeatedly but instead has to understand something, coherence goes down. For example, take an algorithm that can generate images of objects. If the algorithm is restricted to birds, it could do a recognizable bird. If this same algorithm is asked to generate images of any animal, if its task is that broad, the bird it generates becomes an unrecognizable brown feathered smear against a green background.
Spectrum: That sounds… disturbing.
Shane: It’s disturbing in a weird amusing way. What’s really disturbing is the humans it generates. It hasn’t seen them enough times to have a good representation, so you end up with an amorphous, usually pale-faced thing with way too many orifices. If you asked it to generate an image of a person eating pizza, you’ll have blocks of pizza texture floating around. But if you give that image to an image-recognition algorithm that was trained on that same data set, it will say, “Oh yes, that’s a person eating pizza.”
BACK TO TOP↑ Why overestimating AI is dangerous Spectrum: Do you see it as your role to puncture the AI hype?
Shane: I do see it that way. Not a lot of people are bringing out this side of AI. When I first started posting my results, I’d get people saying, “I don’t understand, this is AI, shouldn’t it be better than this? Why doesn't it understand?” Many of the impressive examples of AI have a really narrow task, or they’ve been set up to hide how little understanding it has. There’s a motivation, especially among people selling products based on AI, to represent the AI as more competent and understanding than it actually is.
Spectrum: If people overestimate the abilities of AI, what risk does that pose?
Shane: I worry when I see people trusting AI with decisions it can’t handle, like hiring decisions or decisions about moderating content. These are really tough tasks for AI to do well on. There are going to be a lot of glitches. I see people saying, “The computer decided this so it must be unbiased, it must be objective.”

“If the algorithm’s task is to replicate human hiring decisions, it’s going to glom onto gender bias and race bias.”
—Janelle Shane, AI Weirdness blogger
That’s another thing I find myself highlighting in the work I’m doing. If the data includes bias, the algorithm will copy that bias. You can’t tell it not to be biased, because it doesn’t understand what bias is. I think that message is an important one for people to understand.
If there’s bias to be found, the algorithm is going to go after it. It’s like, “Thank goodness, finally a signal that’s reliable.” But for a tough problem like: Look at these resumes and decide who’s best for the job. If its task is to replicate human hiring decisions, it’s going to glom onto gender bias and race bias. There’s an example in the book of a hiring algorithm that Amazon was developing that discriminated against women, because the historical data it was trained on had that gender bias.
Spectrum: What are the other downsides of using AI systems that don’t really understand their tasks?
Shane: There is a risk in putting too much trust in AI and not examining its decisions. Another issue is that it can solve the wrong problems, without anyone realizing it. There have been a couple of cases in medicine. For example, there was an algorithm that was trained to recognize things like skin cancer. But instead of recognizing the actual skin condition, it latched onto signals like the markings a surgeon makes on the skin, or a ruler placed there for scale. It was treating those things as a sign of skin cancer. It’s another indication that these algorithms don’t understand what they’re looking at and what the goal really is.
BACK TO TOP↑ Giraffing Spectrum: In your blog, you often have neural nets generate names for things—such as ice cream flavors, paint colors, cats, mushrooms, and types of apples. How do you decide on topics?
Shane: Quite often it’s because someone has written in with an idea or a data set. They’ll say something like, “I’m the MIT librarian and I have a whole list of MIT thesis titles.” That one was delightful. Or they’ll say, “We are a high school robotics team, and we know where there’s a list of robotics team names.” It’s fun to peek into a different world. I have to be careful that I’m not making fun of the naming conventions in the field. But there’s a lot of humor simply in the neural net’s complete failure to understand. Puns in particular—it really struggles with puns.
Spectrum: Your blog is quite absurd, but it strikes me that machine learning is often absurd in itself. Can you explain the concept of giraffing?
Shane: This concept was originally introduced by [internet security expert] Melissa Elliott. She proposed this phrase as a way to describe the algorithms’ tendency to see giraffes way more often than would be likely in the real world. She posted a whole bunch of examples, like a photo of an empty field in which an image-recognition algorithm has confidently reported that there are giraffes. Why does it think giraffes are present so often when they’re actually really rare? Because they’re trained on data sets from online. People tend to say, “Hey look, a giraffe!” And then take a photo and share it. They don’t do that so often when they see an empty field with rocks.
There’s also a chatbot that has a delightful quirk. If you show it some photo and ask it how many giraffes are in the picture, it will always answer with some non zero number. This quirk comes from the way the training data was generated: These were questions asked and answered by humans online. People tended not to ask the question “How many giraffes are there?” when the answer was zero. So you can show it a picture of someone holding a Wii remote. If you ask it how many giraffes are in the picture, it will say two.
BACK TO TOP↑ Machine and human creativity Spectrum: AI can be absurd, and maybe also creative. But you make the point that AI art projects are really human-AI collaborations: Collecting the data set, training the algorithm, and curating the output are all artistic acts on the part of the human. Do you see your work as a human-AI art project?
Shane: Yes, I think there is artistic intent in my work; you could call it literary or visual. It’s not so interesting to just take a pre-trained algorithm that’s been trained on utilitarian data, and tell it to generate a bunch of stuff. Even if the algorithm isn’t one that I’ve trained myself, I think about, what is it doing that’s interesting, what kind of story can I tell around it, and what do I want to show people.

The Halloween costume algorithm “was able to draw on its knowledge of which words are related to suggest things like sexy barnacle.”
—Janelle Shane, AI Weirdness blogger
Spectrum: For the past three years you’ve been getting neural nets to generate ideas for Halloween costumes. As language models have gotten dramatically better over the past three years, are the costume suggestions getting less absurd?
Shane: Yes. Before I would get a lot more nonsense words. This time I got phrases that were related to real things in the data set. I don’t believe the training data had the words Flying Dutchman or barnacle. But it was able to draw on its knowledge of which words are related to suggest things like sexy barnacle and sexy Flying Dutchman.
Spectrum: This year, I saw on Twitter that someone made the gothy giraffe costume happen. Would you ever dress up for Halloween in a costume that the neural net suggested?
Shane: I think that would be fun. But there would be some challenges. I would love to go as the sexy Flying Dutchman. But my ambition may constrict me to do something more like a list of leg parts.
BACK TO TOP↑ Continue reading

Posted in Human Robots

#436184 Why People Demanded Privacy to Confide ...

This is part four of a six-part series on the history of natural language processing.

Between 1964 and 1966, Joseph Weizenbaum, a German American computer scientist at MIT’s artificial intelligence lab, developed the first-ever chatbot [PDF].

While there were already some rudimentary digital language generators in existence—programs that could spit out somewhat coherent lines of text—Weizenbaum’s program was the first designed explicitly for interactions with humans. The user could type in some statement or set of statements in their normal language, press enter, and receive a response from the machine. As Weizenbaum explained, his program made “certain kinds of natural-language conversation between man and computer possible.”

He named the program Eliza after Eliza Doolittle, the working-class hero of George Bernard Shaw’s Pygmalion who learns how to talk with an upper-class accent. The new Eliza was written for the 36-bit IBM 7094, an early transistorized mainframe computer, in a programming language that Weizenbaum developed called MAD-SLIP.

Because computer time was a valuable resource, Eliza could only be run via a time-sharing system; the user interacted with the program remotely via an electric typewriter and printer. When the user typed in a sentence and pressed enter, a message was sent to the mainframe computer. Eliza scanned the message for the presence of a keyword and used it in a new sentence to form a response that was sent back, printed out, and read by the user.

To encourage ongoing dialogue, Weizenbaum designed Eliza to simulate the type of conversational style used by a Rogerian psychoanalyst. The program would take something the user said and reflect it back with a question, as can be seen in the example below. (Notice how the program takes keywords like “boyfriend” and “depressed” and re-uses them.)

Human: Men are all alike.

Computer: IN WHAT WAY?

Human: They're always bugging us about something or other.

Computer: CAN YOU THINK OF A SPECIFIC EXAMPLE?

Human: Well, my boyfriend made me come here.

Computer: YOUR BOYFRIEND MADE YOU COME HERE

Human: He says I'm depressed much of the time.

Computer: I AM SORRY TO HEAR YOU ARE DEPRESSED

Human: It's true. I am unhappy.

Weizenbaum chose this mode of dialogue for Eliza because it gave the impression that the computer understood what was being said without having to offer anything new to the conversation. It created the illusion of comprehension and engagement in a mere 200 lines of code.

To test Eliza’s capacity to engage an interlocutor, Weizenbaum invited students and colleagues into his office and let them chat with the machine while he looked on. He noticed, with some concern, that during their brief interactions with Eliza, many users began forming emotional attachments to the algorithm. They would open up to the machine and confess problems they were facing in their lives and relationships.

During their brief interactions with Eliza, many users began forming emotional attachments to the algorithm.

Even more surprising was that this sense of intimacy persisted even after Weizenbaum described how the machine worked and explained that it didn’t really understand anything that was being said. Weizenbaum was most troubled when his secretary, who had watched him build the program from scratch over many months, insisted that he leave the room so she could talk to Eliza in private.

For Weizenbaum, this experiment with Eliza made him question an idea that Alan Turing had proposed in 1950 about machine intelligence. In his paper, entitled “Computing Machinery and Intelligence,” Turing suggested that if a computer could conduct a convincingly human conversation in text, one could assume it was intelligent—an idea that became the basis of the famous Turing Test.

But Eliza demonstrated that convincing communication between a human and a machine could take place even if comprehension only flowed from one side: The simulation of intelligence, rather than intelligence itself, was enough to fool people. Weizenbaum called this the Eliza effect, and believed it was a type of “delusional thinking” that humanity would collectively suffer from in the digital age. This insight was a profound shock for Weizenbaum, and one that came to define his intellectual trajectory over the next decade.

The simulation of intelligence, rather than intelligence itself, was enough to fool people.

In 1976, he published Computing Power and Human Reason: From Judgment to Calculation [PDF], which offered a long meditation on why people are willing to believe that a simple machine might be able to understand their complex human emotions.

In this book, he argues that the Eliza effect signifies a broader pathology afflicting “modern man.” In a world conquered by science, technology, and capitalism, people had grown accustomed to viewing themselves as isolated cogs in a large and uncaring machine. In such a diminished social world, Weizenbaum reasoned, people had grown so desperate for connection that they put aside their reason and judgment in order to believe that a program could care about their problems.

Weizenbaum spent the rest of his life developing this humanistic critique of artificial intelligence and digital technology. His mission was to remind people that their machines were not as smart as they were often said to be. And that even though it sometimes appeared as though they could talk, they were never really listening.

This is the fourth installment of a six-part series on the history of natural language processing. Last week’s post described Andrey Markov and Claude Shannon’s painstaking efforts to create statistical models of language for text generation. Come back next Monday for part five, “In 2016, Microsoft’s Racist Chatbot Revealed the Dangers of Conversation.”

You can also check out our prior series on the untold history of AI. Continue reading

Posted in Human Robots

#436176 We’re Making Progress in Explainable ...

Machine learning algorithms are starting to exceed human performance in many narrow and specific domains, such as image recognition and certain types of medical diagnoses. They’re also rapidly improving in more complex domains such as generating eerily human-like text. We increasingly rely on machine learning algorithms to make decisions on a wide range of topics, from what we collectively spend billions of hours watching to who gets the job.

But machine learning algorithms cannot explain the decisions they make.
How can we justify putting these systems in charge of decisions that affect people’s lives if we don’t understand how they’re arriving at those decisions?

This desire to get more than raw numbers from machine learning algorithms has led to a renewed focus on explainable AI: algorithms that can make a decision or take an action, and tell you the reasons behind it.

What Makes You Say That?
In some circumstances, you can see a road to explainable AI already. Take OpenAI’s GTP-2 model, or IBM’s Project Debater. Both of these generate text based on a large corpus of training data, and try to make it as relevant as possible to the prompt that’s given. If these models were also able to provide a quick run-down of the top few sources in that corpus of training data they were drawing information from, it may be easier to understand where the “argument” (or poetic essay about unicorns) was coming from.

This is similar to the approach Google is now looking at for its image classifiers. Many algorithms are more sensitive to textures and the relationship between adjacent pixels in an image, rather than recognizing objects by their outlines as humans do. This leads to strange results: some algorithms can happily identify a totally scrambled image of a polar bear, but not a polar bear silhouette.

Previous attempts to make image classifiers explainable relied on significance mapping. In this method, the algorithm would highlight the areas of the image that contributed the most statistical weight to making the decision. This is usually determined by changing groups of pixels in the image and seeing which contribute to the biggest change in the algorithm’s impression of what the image is. For example, if the algorithm is trying to recognize a stop sign, changing the background is unlikely to be as important as changing the sign.

Google’s new approach changes the way that its algorithm recognizes objects, by examining them at several different resolutions and searching for matches to different “sub-objects” within the main object. You or I might recognize an ambulance from its flashing lights, its tires, and its logo; we might zoom in on the basketball held by an NBA player to deduce their occupation, and so on. By linking the overall categorization of an image to these “concepts,” the algorithm can explain its decision: I categorized this as a cat because of its tail and whiskers.

Even in this experiment, though, the “psychology” of the algorithm in decision-making is counter-intuitive. For example, in the basketball case, the most important factor in making the decision was actually the player’s jerseys rather than the basketball.

Can You Explain What You Don’t Understand?
While it may seem trivial, the conflict here is a fundamental one in approaches to artificial intelligence. Namely, how far can you get with mere statistical associations between huge sets of data, and how much do you need to introduce abstract concepts for real intelligence to arise?

At one end of the spectrum, Good Old-Fashioned AI or GOFAI dreamed up machines that would be entirely based on symbolic logic. The machine would be hard-coded with the concept of a dog, a flower, cars, and so forth, alongside all of the symbolic “rules” which we internalize, allowing us to distinguish between dogs, flowers, and cars. (You can imagine a similar approach to a conversational AI would teach it words and strict grammatical structures from the top down, rather than “learning” languages from statistical associations between letters and words in training data, as GPT-2 broadly does.)

Such a system would be able to explain itself, because it would deal in high-level, human-understandable concepts. The equation is closer to: “ball” + “stitches” + “white” = “baseball”, rather than a set of millions of numbers linking various pathways together. There are elements of GOFAI in Google’s new approach to explaining its image recognition: the new algorithm can recognize objects based on the sub-objects they contain. To do this, it requires at least a rudimentary understanding of what those sub-objects look like, and the rules that link objects to sub-objects, such as “cats have whiskers.”

The issue, of course, is the—maybe impossible—labor-intensive task of defining all these symbolic concepts and every conceivable rule that could possibly link them together by hand. The difficulty of creating systems like this, which could handle the “combinatorial explosion” present in reality, helped to lead to the first AI winter.

Meanwhile, neural networks rely on training themselves on vast sets of data. Without the “labeling” of supervised learning, this process might bear no relation to any concepts a human could understand (and therefore be utterly inexplicable).

Somewhere between these two, hope explainable AI enthusiasts, is a happy medium that can crunch colossal amounts of data, giving us all of the benefits that recent, neural-network AI has bestowed, while showing its working in terms that humans can understand.

Image Credit: Image by Seanbatty from Pixabay Continue reading

Posted in Human Robots