Artificial intelligence as science has existed since 1956, although the concept of a thinking machine has been around for centuries. The first work that has been recognized as AI, artificial neurons, was created a few years earlier in 1943 by McCullouch and Pitts. Many have since predicted that in a generation, the human race will create a machine that can do the work of humans. Consistently, though, the target has been moving ever forward. Why would it be different now?
Despite all the hype and talk about the singularity (the point when computers will be more intelligent than humans), we still lack some very basic features in AI, before it can be considered a real artificial intelligence that can rival human capabilities.
Let’s look at the different waves of AI:
- 1st Wave – decision tree or symbolic AI: With development of computing power in 1980’s, computer systems that had a structured input and binary logic based “brain” were able to rapidly and reliably resolve narrow and predefined problem areas. They however faced problems with anything outside of their programmed field, and were not capable of learning or autonomous operation.
- 2nd Wave – deep learning: When neural networks were discovered in 1940-1950, the models were used for pattern recognition but required such processing power that the development of deep networks (multiple combined neural networks) stalled. With powerful graphic processors and scientific advances from 2006 onwards, the developed neural networks are capable of human-like learning and perception, however they are still not capable of autonomous reasoning without significant help from a human to tell if the network produced a correct result.
- Autonomous AI: A “real” artificial intelligence is considered to be able to learn like a human baby, by observing and experimenting. There are some companies in the market who are developing an AI with all four needed skills (perception, learning, reasoning, autonomy), however the computing power required to utilize these capabilities in any meaningful way is still 3-5 years away.
The current hype wave is riding on the 2nd wave, and is starting to realize already some wonderful, albeit limited results. Deep learning is already successfully demonstrated in speech (and other pattern) recognition, data processing and control mechanisms, where sufficiently large datasets of sufficiently limited problems can be found. However the current solutions are limited in the amount of training required (especially in areas with wide problem scope) and considerable hardware requirements.
So what should we do next?
Although many companies may be anxious to jump to the AI train and see where the ride takes them, the trip should be planned meticulously. Here are three things that organizations should currently consider before any investments into AI:
- Data collection: Both waves 2 and 3 require extraordinary amounts of data in order to learn patterns. However for wave 3, especially in wide problem areas, the data amounts are considerably more than is required for current AI solutions. Hence, whatever else is in your AI plan, make sure to invest considerably in collecting large amounts of data, especially focusing on discussions, internal decision making and “silent knowledge”. Naturally, the collected and available data needs to be correct and usable.
- Impact: When considering any of the solutions above, a holistic view to the whole process network is needed. There are often possibilities to automate and simplify longer stretches of the process, if only organizations could analyze variations and impacts of the process outcomes. In addition, each investment should come with a clear (yet simple) business case or business model to ensure the investment delivers value
- Change readiness: Not many people are willing to hand over their tasks to an AI, although it probably does the work many times better. While considering the AI transformation in the organization, it’s worth to also consider how to get people involved and engaged with the new digital co-workers. Starting with small successes, communicating and involving people will ensure that value can be sustainably created by the AI solutions.
You may have noticed that Amazon Alexa and Google home want to position their smart home controls so that they can listen in to your discussions decision making and behavior (record, analyze and store),. They may just be preparing for wave 3 – how will you do the same?