Waves of AI
THE TERM ‘ARTIFICIAL INTELLIGENCE” was coined in 1956, at a historic conference at Dartmouth, and in the recent years we see glimpses of its transformational power and capabilities. AI is proving to be a general purpose technology (GPT) that changes the way we work, life and play. Like electricity and the computer before it, AI can be applied to every industry and an almost infinite set of problems. Technology companies and nations are racing for supremacy.
It’s difficult to gain consensus on the definition of AI but everyone agrees its an area of computer science, a field of science and primarily computer science. Like the Russian nesting doll, AI is the parent field, with subfields like machine learning, deep learning, NLP and more.
Zachary Lipton, Assistant Professor of Computer Science at Carnegie Mellon University describes the term AI as aspirational, a moving target based on those capabilities that human posses that machines do not. This easily explains the waves as our aspirations for AI are constantly increasing, technologies change and emerge.
Media pundits and almost everyone use the word, AI, liberally, as a broad category describing solutions, apps, or products that use NLP, Deep Learning, computer vision, machine learning or other technologies. Vendors jump on the AI bandwagon to sell their wares and companies label and tag their products as AI for a variety of reasons. But what does AI mean when we describe something as AI? Why should we care?
We should care because it advances the field of study when we understand how problems are solved, the techniques, architecture, algorithms and / or tools. It also helps buyers understand what they are buying. In a worse case scenario wouldn’t you rather have a solution that gets better and better with more data than one that requires code refactoring as the real world changes? A solution that does not carry the same technical debt or is future proof? A more agile solution? This requires we understand the specific AI technologies being used. For example, Is NLP being used or NLP with deep learning?
There is no right or wrong answer and it’s okay to label all approaches AI but it is useful to know the how (the specific AI methods, technologies, paradigms, architecture or approaches). Such an understanding provides awareness of the AI solution’s agility and scale; it’s ability to adapt and improve over time, and its present and future capabilities. Hence the conversation on AI Waves.
The conflation of AI with machine learning or analytics further exacerbates the distinction and value propositions of specific deep learning architectures, paradigms and disciplines. AI is changing software development whereas even the mantra of “fail fast” customary of modern software engineering has its limitations with AI development. AI development sometimes requires a blend of creativity and continuous learning, often there is no “failure” but simply an epiphany or discovery leading to more novel approaches for problem solving.
Classical AI - Wave 0
Much of the 20th century work in AI can be lumped into this category. I categorize intelligent computers or systems predominantly with procedure code, developers writing code fits into this category. Although elements of machine learning may be at play this wave is characterized with procedural programming logic as the main programming technique for construction. The iconic display of Wave 0 may have been IBM’s Deep Blue's computer beating one of humanity's great intellectual champions in chess, Gary Kasparov. At the time this was heralded as a sign artificial intelligence was catching up to human intelligence. Although years later Centaur Chess, https://www.huffpost.com/entry/centaur-chess-shows-power_b_6383606, would show the power of teaming humans and machines.
Rise of Machine Learning and NLP - Wave 1
In 2011, IBM again made an iconic display of AI, the IBM Watson computer system competed on Jeopardy! against legendary champions Brad Rutter and Ken Jennings and won. This 2011 IBM Watson system took several years to build, brute force in the hardware implementation, a lot of math, machine learning and NLP. No graphical processor units (GPUs) were used nor was deep learning used, both characteristics of Wave 2 systems, prompting criticism from some at Davos in 2016,
To make it very clear, some of the so-called disruptions of the big players like IBM or SAP is not real. IBM’s Watson is not disruptive AI, it’s a rehash of simple algorithms known since the eighties -– and stuff like Deep Mind from Google, Scaled Inference, and Vicarious are knocking its socks off. … Big companies are saying: “Have disruption problems? Oh, just buy our rebranded, legacy software and you’ll be safe.” This is dangerous.
To the casual viewer it may have seemed that IBM Watson on Jeopardy! was doing conversational AI and that was not the case as ASCII files were transmitted.
As data became more abundant, easier to gather, a data driven and probability-based approach dominated, ushering in the rise of analytics and machine learning, Wave 1.
Rise of Deep learning - Wave 2
Contrary to the opinions of some what we can do with AI today is different than what we could do in the 1950s, 1960s, 1970s, 1980s and 1990s and early 2000s. Although we are using many of the same algorithms or computer science, its different today. Deep learning, new algorithms (e.g.,, back propagation), GPUs and massive amounts of data is the difference. There has been an explosion of new thinking by industry and academia, there is clearly a renaissance going on today. Anyone participating in Kaggle competitions, reading the flood of research from universities, playing with the open source software, watching the almost daily AI announcements from the big technology companies, or just paying attention to startups and the Internet / Cloud era companies see the impact of Wave 2. The old school NLP aka circa Watson 2011 has been thrown out of the window.
In October 2015, the original AlphaGo became the first computer system to beat a human professional Go player without a handicap. In 2017, the successor, Alpha Go Master, beat the world No.1 ranked player at the time in a three gae match. AlphaGo uses deep learning and is the iconic display of Wave 2. Deep learning and all of its subfield is the defining architecture and technology of Wave 2.
The demonstration of neural nets, deep learning algorithms, reaching human perception in sight, sound, and reading set in motion a tsunami of AI research and development in academia and technology companies and brought about Wave 2. The Internet provided the mother lode of data and GPUs the compute power for crunching.
Correlation to Causality - Wave 3
Daniel Pearl best describes what’s next in his groundbreaking book, “The Book of Why, the New Science of Cause and Effect.” Turing Award recipient Yoshua Bengio speaking at an international assembly of AI researchers in Vancouver this year (2019), provides some specifics, see https://journalismai.com/2019/12/12/yoshua-bengio-from-system-1-deep-learning-to-system-2-deep-learning-neurips-2019/
Summary
A shift from rules based to a data driven approach paved the way for a rise in a new era in AI. The Internet provided the mother lode of data and GPUs the compute power for crunching. The demonstration of neural nets and deep learning algorithms, reaching human perception in sight, sound, and reading set in motion a tsunami of AI research and development in academia and technology companies bringing about Wave 2.
Artificial intelligence will become a differentiating and disruptive factor in all industries. Artificial intelligence (AI) isn't an end in itself, but a strategic capability that companies must mature and democratize to achieve measurable and transformative business outcomes. Company’s ability to maximize the promise of AI will be fulfilled only by understanding AI is different now, people still matter, the old rules may not apply, software engineering will be different and we can reimagine the future of how work is done.