by Peter B. Nichol

Breaking down artificial intelligence to form a starting point for adoption

Aug 14, 2017
AnalyticsArtificial IntelligenceConsumer Electronics

To leverage, communicate and sell the power of artificial intelligence, we first must capture its essence.

conceptual artificial intelligence
Credit: Thinkstock

Artificial intelligence will humanize recommendation engines, improve the accuracy of logistics engines, and represent a monumental change in the friendliness of chatbot engines. Learning new languages (Duolingo), finding new dinner plans (Replika) and making photography exciting again (Prisma) is how our business partners will be introduced to the potential of artificial intelligence.

How we plan for AI

If I asked you how to build a house, you’d have a series of steps in mind. When asked how to validate a company’s technology security perimeter, other action steps come immediately to the forefront. And when booking a vacation to Brazil, a clear approach to get you on the beach fast rushes to the mind.

We’re of course not talking about building houses, creating security resilience, or booking vacations. We’re talking about how to introduce business leaders, scientists and medical professionals to the power of artificial intelligence. So where do we start? What’s our first step?

Three steps toward AI enlightenment

We start with a framework for all intelligence agents. Artificial intelligence can be separated into two categories: (1) thought processes and reasoning and (2) behavior. Whether you lean more toward the mathematics and engineering side (rationalist) or closer to the human-centered approach (behavior), the heart of AI is trying to understand how we think.

The first step: Decide which of the four categories of artificial intelligence the enterprise will explore.

  1. Thinking humanly: systems that think like humans
  2. Acting humanly: systems that act like humans
  3. Thinking rationally: systems that think rationally
  4. Acting rationally: systems that act rationally

The second step: determine the intent of our artificial intelligence initiative.

Thinking humanly (cognitive modeling) blends artificial intelligence with models—as in the case of neurophysiological experiments. Actual experiments in the cognitive sciences depend on human or animal observations and investigations. Acting humanly (Turning Test) attempts to establish a line between non-intelligence and satisfactory intelligence. Thinking rationally captures “right thinking” in computer language. Coding logic is fraught with challenges, since informal knowledge doesn’t translate well to formal notation. Acting rationally is about acting. Agents perform acts, and “rational agents” can autonomously maneuver, adapt to change and evolve (learned intelligence).

The third step: identify the capabilities required.

Thinking humanly capabilities:

  1. Observation
  2. Matching human behavior
  3. Reasoning approach to solving problems
  4. Solve problems
  5. Computer models to simulate the human mind

Acting humanly capabilities:

  1. Natural language processing
  2. Automated reasoning
  3. Machine learning
  4. Knowledge representation
  5. Computer vision and robotics

Thinking rationally capabilities:

  1. Codify thinking
  2. Pattern argument structures
  3. Codify facts and logic (knowledge)
  4. Solve problems in practice (not principle)
  5. Solve problems with logical notation

Acting rationally:

  1. Thought inferences
  2. Adapt to change (agents, chatbots)
  3. Analyze multiple correct outcomes
  4. Operate autonomously
  5. Create and pursue objectives

Step beyond

Artificial intelligence, since the mid-1940s, has moved across the plane of learning from philosophy to control theory. The philosophy of logic and reason established the foundations of learning, language and rationality. Mathematics formally represented computations and probabilities. Psychology illuminated the phenomena of motion and psychophysics (experimental techniques). Linguistics studied morphology, syntax, phonetics and semantics. Neuroscience poked at the function of the nervous system and brain. Control theory combines the complexities of dynamic systems and how behavior is modified by feedback.

Navigation, neural networks, gene expression, climate modeling and production theory all stem from control systems engineering.

It’s easy to become tangled up in the possibilities of artificial intelligence. First, we must decide which of the four categories of artificial intelligence we will explore. Second, we must determine the intent of our artificial intelligence initiative. Third, we must identify the capabilities required. In sum: Start with a plan and clarify your first three steps for your organization to realize the potential of artificial intelligence.