Gmail now completes sentences. When a user types "Thanks for your," it suggests "email" in grey text, and it is usually correct. Smart Compose was introduced last year, and we have been using it for months without examining how it works.
Last week, DeepMind demonstrated a program called AlphaStar defeating professional StarCraft II players. That result drew the headlines. The grey text in the inbox, however, is the more relevant development for businesses.
The AI that is the subject of so much commentary is not a future development. It is already present in the tools businesses use every day, and most users do not understand how it works. This year, we have set out to address that gap.
What is machine learning, in plain words?
Conventional software follows rules written by a person: if an email contains a given word, place it in a given folder. Machine learning reverses this process. The software receives a very large set of examples and derives the rules itself.
A system shown millions of emails that people marked as spam becomes effective at detecting spam. A system shown millions of sentences people typed after "Thanks for your" becomes effective at predicting the next word.
The system does not understand what spam is. It has identified a pattern.
This distinction matters more than it may seem. A pattern learned from examples is only as reliable as those examples. If the examples are outdated, unbalanced or simply wrong, the system learns those flaws as well and then acts on them with complete confidence. Vendors of these tools rarely highlight this limitation.
Where is AI already present in the tools small businesses use?
It appears throughout an ordinary workday, although none of these tools is labelled "AI":
- The spam filter that keeps a booking inbox usable.
- Google Maps estimating a customer's driving time based on traffic.
- Google Translate, which moved to a neural system a few years ago and became considerably more accurate.
- Google Photos finding "boat" or "turtle" in thousands of pictures that were never tagged.
- Search itself; Google has used machine learning to rank results for some time.
- Facebook and Instagram determining which of your posts people actually see.
The last item has the most direct financial impact on a business. A dive centre on a small island depends on visibility, and the system that determines that visibility is a learning system that no one outside Facebook can examine. We devote considerable time to understanding what it favours, as does everyone who manages a small business page.
Businesses are therefore not choosing whether to use AI; they are already being ranked by it.
This is the appropriate starting point for a small business owner. Rather than focusing on broader debates about AI, identify where these tools already affect your customers and your work.
Why learn the basics before the hype peaks?
We have experienced several such waves. At Raincity Studios in Vancouver, we went through the blog boom and then the social media boom, and we observed many businesses spend money on technologies they did not understand because a presenter advised them to. The businesses that fared well were those that took time to understand the technology themselves before it became a sales pitch.
We expect AI to follow the same path, on a larger scale. Consultants are already positioning themselves, and the label "AI-powered" is beginning to appear on products that, as far as we can determine, consist of a few if-then rules and an attractive logo.
Our approach is therefore straightforward: learn enough now, while the conversation is still largely among engineers, to evaluate claims accurately later.
Which free course do we recommend?
The best free resource we have found is Elements of AI, an online course from the University of Helsinki and Reaktor. It is written for a general audience. It requires no coding and no advanced mathematics, and it is candid about what AI cannot do, which is uncommon. Finland set out to teach the basics to a significant share of its population, an approach we find far more constructive than much of the excited coverage in the news feed.
We have completed several chapters. Progress is gradual on a dive island with intermittent internet access, but it is the first resource we have found that addresses readers as neither programmers nor beginners without background.
What do we recommend to a small business owner now?
We began studying this subject in a structured way a few weeks ago, and several recommendations are already clear.
List the tools you use and identify the features that make a prediction for you: suggested replies, ad audiences, the order in which reviews are displayed, translation, the photo app that sorts your pictures. These are the AI systems you already use. Understand how they behave before purchasing any new product that carries the label.
A pattern is not a fact. When a tool makes a suggestion, it is making a prediction based on the examples it has seen. The prediction is often good, but sometimes it is wrong while appearing entirely confident. The decision remains yours.
Observe what these tools get wrong. Running a menu or a dive briefing through Google Translate will show within five minutes where these systems perform well and where they fail. If that output reaches a customer without review, your business is accountable for it, not Google.
Finally, invest an evening or two. There is no need to become an engineer, only to learn enough that the next vendor offering an "AI solution" cannot mislead you. Accessible AI training for small business owners is still very rare, and we expect the need for it to arise much sooner than most people anticipate.
Our assessment is that AI for writing and conversation is further away than the headlines suggest, while the discreet sorting and prediction functions are much closer than most people realize. Both developments remain uncertain, but early preparation is the prudent course.