All technology is hard to implement at the start. New technology often requires specialized personnel to manage its intricacies.
As the technology progresses, one of two things will happen: Either it will remain difficult (and expensive) to implement or abstraction layers will be added, and the technology will become easy enough for a more general practitioner. If the first case occurs, the technology will remain an expensive niche or, potentially, die out. If the second scenario happens, the technology will become widely adopted and may evolve to the point that a general business user can make use of it without the assistance of the technical staff.
This is the point the market has reached with AI. Either AI will become highly specialized and restricted to a few use cases or it will become democratized. For democratization of AI to take hold, a big problem must be addressed — the degree of difficulty when developing AI applications.
AI Is in High Demand, But Related Skills Are Scarce
AI is about prediction not cognition. Computers do not “understand” the way humans do. Instead, they interpret data based on statistical models that calculate probabilities and determine a most likely path. They are not creative, cannot take risks, nor can they develop their own models. Humans do some of this intuitively but can also develop schemas (human models of the world) based on very little data and evolve them quickly. Computers can refine their outputs based on more data, adjusting probabilities over time, but cannot radically alter their human-created models.
Even machine learning is not the same as human cognition. It can discern a function — a kind of a model — from data and use that to predict outcomes. Those predications, however, are based on statistical algorithms. In other words, math that most of us don’t use every day.
The skills required to do this work — programming, math, data analysis and domain knowledge — are expensive. They are expensive because they are rare and require an enormous amount of education. Typical AI projects require specially trained data scientists and people with doctorates to develop at least some of the code. Skills and education of this sort are not a common IT skill set. Thus, the barrier to entry for AI applications is still high. That barrier makes AI inaccessible to most IT departments that lack the budget to hire PhDs or consultants with these skills.
For AI to appeal to a wider constituency, it will need to become more like other parts of IT software systems. AI will need to leverage existing skills, use common IT tools, and eventually evolve to be usable by non-technical people.
Related Article: 5 Tips for Attracting AI Talent
What It Will Take to Democratize AI
Some of what must happen for the democratization of AI to occur is:
- Accessing AI generated data using SQL. Ultimately, AI models generate data that has to be dealt with programmatically. The most common method of accessing data in IT systems is using SQL. This method makes sense for accessing result sets from AI models.
- Templates and blackbox operation. Many applications, within a problem domain, could use similar models. Instead of each company developing their own models, it would be much easier if there were templates that could be adjusted to a company’s needs. Even better would be blackbox models that would only need to be trained and tweaked. It would be even more useful if these standard models were accessible from a GUI. This would allow a domain expert to manipulate variables in order to customize the model without technical assistance.
- Ultimately, AI will have to evolve into a low-code, no-code form that is similar to most common business applications. This would eliminate the need to have specifically skilled AI experts involved in application development.
Thankfully, much of this exists in some form or another. They are mostly available in a piecemeal form, with some vendors supporting one of these characteristics or features and others another one. Eventually, AI should coalesce around the majority of these features, making AI capable of being used by a majority of companies.
Most IT follows this path from science to democratization or else becomes a niche or dies. AI is showing signs of being on the path to democratization and mass adoption. It still has the potential to fail, but we are almost past the inflection point where AI becomes a common business tool.
Related Article: Is Your Enterprise Ready for 'Democratized AI'?