“Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child's? If this were then subjected to an appropriate course of education one would obtain the adult brain.”
“Can machines think?"... The new form of the problem can be described in terms of a game which we call the 'imitation game." It is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart front the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either "X is A and Y is B" or "X is B and Y is A." The interrogator is allowed to put questions to A and B... We now ask the question, "What will happen when a machine takes the part of A in this game?" Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, "Can machines think?”
Alan Turing, Computing machinery and intelligence (1950)
The Need for Connecting The Dots
I wanted to develop a systems map that focused on charting out the advancements in Artificial Intelligence (AI). In brief, I sought to create something that displayed a historical and social understanding of the processes and methods developed within the field while being mindful of human networks and their corresponding gender, institutional and geographic distribution, as well as the overarching techniques that they create or advance. All the while, where ever possible, I also wanted to tease out an appealing connective thread between various research papers in order to have a more universal and accessible framework.
Games Driving Advances in AI
Bruce G. Buchanan 2006 A (Very) Brief History of Artificial Intelligence gave me ideas for two potential methods with which I could organize my research;
- I could look at how technology or AI in this sense could be described as an extension of man (similar to Ernst Kapp and Marshall McLuhan), which may prove to be a useful organization method for a large set of diverse papers, or ;
- I could focus on the ‘Games’ metaphor which would allow me to place AI in a bigger historical context by looking at the chess playing Mechanical Turk as a precursor to Alan Turing's imitation game challenge. The latter having a visibly lingering effect on the continuing development of AI/Machine Learning (i.e. DeepMind’s AlphaGo/Zero and Dota 2 bots.).
Ultimately, I found the humans vs AI paradigm demonstrated in game challenges to be a more accessible thread due to its appeal to both AI experts and a general audience.
The Metadata
The metadata used in the organization of the system this research produced are: publication dates, author names, gender, related institutions and geographical locations of said institutions, fields that intersect the research, either through the authors’ prior educational experiences or based on the papers cited, industries where the paper has had or could have an impact, and finally the direct output/result of the paper. While the last four categories above can be described extensively, I have made an attempt at limiting the boundaries to perceived focal points in each one of them.
A Note on the Deficiencies of the Metadata
Before looking at the system’s components and how they help us understand and complement AI and other related systems, it’s best to look at the deficiencies of the metadata. Some difficulties in categorizing advancements in the field that are likely to be applied in many industries have been placed under “Industry Research” (for example papers coming out of IBM Research.) Other limitations also appear when looking at the “approaches” category, mostly due to my own limitations in understanding the ever evolving sophistication of machine learning techniques. Some terms might be redundant, and even where terms like “heuristic searches” might not be visible in more recent papers, readings show that heuristic methodologies still underpin decision models in cases like AlphaGo and Jeopardy! Watson.
The academic fields are also limited in that they account mostly for the authors' accredited degrees that I uncovered in the research process, but lack the depth that could capture their expanded interests. In my goal to understand the social elements of the AI fields, it is also perhaps more problematic that categories such as gender remain binary for now, which requires a revisit if this system was to be used extensively. Institutions and geographies pose a similar problem as I was forced to limit my search to places where the authors’ completed their graduate level studies (ignoring nationalities.) This makes for a manageable boundary, but it’s evident that if the institutions are to serve in mapping influential social networks and relationships, the extent to which these could evolve informally is ignored in this process.
Having said that, the system mapping as is does provide some useful cases in understanding the theoretical and practical relationships. By looking at the 10 research papers or publications that have been classified here, we begin to see the influences of early thinkers on Artificial Intelligence, most notably Alan Turing’s Computing machinery and Intelligence, in using games and human challengers as means of developing tactics and strategies that can be replicated in more general problems. This is captured by Turing’s philosophy which also asks whether such achievements can be understood to reflect “thinking” or cognitive capabilities in machines. Indeed, by organizing these human vs machine paradigm based papers in a chronological order, we can begin to associate early agents in checkers and chess as precursors to the more recent iterations that have resulted in the solving of these games. By extending this approach, we can also look at tangential studies such as Luhn’s information retrieval methods as being a precursor to Jeopardy! Watson’s much more advanced ability to convert enormous amounts of data into useful information in a natural language context. Finally, Stanley, the DARPA Grand Challenge winning autonomous car from Stanford, is contextualized here as the beginning of a human vs machine “competition” centered around driving/racing.
The games within the map follow a chronological order that is based on the games being solved or a human opponent being defeated decisively. This order gives us checkers, chess, Stanley/driving, Jeopardy, AlphaGo and Poker from left to right. The research papers that have backed the achievements appear numbered and cited on the legend (far right.)Looking in from the outside are three male and one female characters. The duplicated male characters represent the homogenous nature of the field when it comes to people involved. The woman is trying to take a look, as she is ignored or pushed aside. There’s also a poster of the Mechanical Turk as a fun reference to historic attempts to create thinking machines. Behind it all is a sort of Easter Egg, a fitting wallpaper for our growing AI, that is generated by Martin Thoma using DeepDream, the computer vision program developed by Google which uses convolutional neural networks to create these images.
The Final System – A Knowledge Map
I had a fairly broad audience in mind; people who might have come across news of AI developments and human-computer/machine duels. They would be familiar with some of the more popular ones like Jeopardy! Watson or AlphaGo.
Initially as I was considering using the “technology as the extension of man” metaphor to visualize AI advancements as body parts. As mentioned previously, I settled on a system that was guided by Alan Turing’s “Computer Machinery and Intelligence” which helped inspire this final visual map. I took Turing’s earlier work and theory of developing a program that “simulates the child’s [mind].” I pushed this idea visually by setting the systems map within a child’s bedroom, with the toys (or games) laid out across the room. I also had to reduce how much of each metadata would appear in the poster so as to manage the image density. I believe I was able to capture most of it, though in some cases I could only hint at them (ex: industry category- hint at military with WarGames computer program appearing on the computer screen.)
A poster of Alan Turing at the age of 16 starts us off on the left side of the map. Directly above it are pennants representing the many academic institutions that have contributed to AI research either directly or have appeared through the authors’ works. Companies like IBM and Bell Technologies are presented as stickers on the bookshelf. The bookshelf itself carries many books, most of which are titled to reflect the different fields that have converged or interacted to achieve some of the milestones reflected in the games. Some are left to be filled in the future.
In my research I found that AI research based on games sits within a larger system. The outcomes that have led to the achievements of the machine/program are themselves amalgamations of many other forms of AI, or in some cases stand as catalysts for further advancements in the field (AlphaGo leading to AlphaGoZero for example.) This means that each one of these games intersects with other sub-sytems of AI such as Text Search (Hans Peter Luhn), Natural Language Processing (Watson), computer vision (Stanley), machine learning (Arthur Samuel’s checkers), Reinforcement Learning (AlphaGo) and so on.
The original "mechanical Turk"
So What?
The system, as captured in the authors, publication dates, institutions and industries, firstly allows us to understand the historical and contemporary relationships of AI advancements within a socio-political framework. While the more recent networks seem to be rather dense as is the case in the sizeable teams involved in Stanley, Jeopardy! Watson and AlphaGo, we can point to the ongoing relevance and even historical centrality of academic and research institutions like Cambridge, Oxford, MIT, Stanford, Carnegie Mellon, University of Alberta, IBM, and Bell Labs. The same applies to governmental agencies like the military backings expanding to government supported research agencies like DARPA. Finally, perhaps not surprisingly, we see a system that has omitted the roles of women or people of color, underlined by the homogenous environment of white male researchers - with a slight increase in diversity more recently.
In many ways, this human vs. machine paradigm goes beyond the field of AI itself. From golem, the mechanical Turk, to Frankenstein, R.U.R and Turing’s test, this remains perhaps AI’s most popularly communicated message. This thread can remain useful in the future because it positions the growth of AI within the field’s own perceived “holy grail”, that of Artificial General Intelligence.