Most of the AI on Your Phone Is Ordinary Software

Most of what gets sold to you as AI is ordinary software with a better advertisement. A smaller set of things in your day really are AI, and those are the ones that deserve your attention, because they are the ones that learn from data and make guesses about you.

That is a position, and you can test it on your own phone in about five minutes.

The question that sorts everything

Ask one thing about any feature: did a person write the rules, or did a system learn a pattern from a huge pile of examples and apply it to you?

A heating schedule that warms the house at 6am follows a rule. A feature that predicts what you want to watch next because of what millions of other viewers did has learned a pattern. The first is software. The second is what most people mean when they say AI.

The line blurs at the edges, and plenty of marketing blurs it on purpose. But the question gets you most of the way, and it changes what you should expect from the thing. Rules do exactly what they were told, every time. Learned systems are right most of the time and confidently wrong the rest.

What usually fails the test

Plenty of things sold as intelligent are not.

• Smart home routines. Lights that switch on at sunset or a thermostat that follows a timetable you set are rules. Some thermostats do learn your habits from past behaviour, so read what yours says it does before you credit it with anything.
• Reminders and calendars. The reminder itself is a rule. The part that understood your spoken request (“remind me when I get home”) is the learned part, which is a useful way to see how one product can contain both.
• Price comparison and “deals”. Sorting products by cost is a sort, not a thought.
• Fitness bands and wearables. They measure things such as heart rate and movement. Measuring is not diagnosing, and a band that claims to spot illness early is making a health claim that belongs with your GP, not with a gadget.

None of this makes these features bad. It just means you do not need to ascribe any judgement to them, and you can keep your scepticism for the places it is needed.

What really passes

A short list of everyday things genuinely run on learned models:

• Spam and scam filtering in email and messages, which learns what unwanted messages tend to look like.
• Fraud flags on card spending, where unusual transactions are picked out by comparing them with patterns of normal behaviour.
• Photo apps that group faces or find every picture containing a dog.
• Speech to text, autocomplete and translation, all trained on enormous amounts of language.
• Map arrival times, which blend live traffic with patterns learned from past trips.
• Recommendation feeds on streaming, shopping and social apps.
• Chat assistants built on large language models, the newest and most talked about of the lot.

Notice what each of these needs from you. Most of them run on your behaviour, your photos, your words or your location. That is the real price of the convenience, and it is why the rest of this piece is about privacy rather than robots.

Where Australian privacy rules sit

The Privacy Act covers personal information whatever technology handles it, so a chat assistant or a photo app is not exempt just because it is new. The Office of the Australian Information Commissioner has published guidance on privacy and the use of commercially available AI products, and its message to organisations is cautious: be careful about putting personal information into publicly available AI tools, and treat sensitive information with particular care.

You can borrow that caution for yourself. Treat a chat window like a postcard. Do not paste in anything you would not want read by someone other than the person you meant it for: medical details, tax file numbers, a customer’s data from work.

Separately, the Australian Government has published a voluntary set of AI ethics principles, covering matters such as privacy protection, fairness, transparency and the ability to contest an outcome. They are not law. They are a useful yardstick for asking a company what it does with your data.

Data can also end up with the wrong people. If you want to see what happens when it leaks, how large data breaches unfold is worth reading alongside this.

The fair objection

The best counter to all of this is simple: if a feature helps me, who cares whether it is technically AI?

For the result, perhaps you do not. But the label changes two things you should care about. A learned system can be wrong in ways a rule cannot, because it generalises from examples and sometimes generalises badly. And it usually runs on your data, which brings back the privacy questions above.

There is also a fair point against my own test: rule-based software can be just as unfair or just as careless with your information as a learned model. The test sorts how a feature works, not whether it is trustworthy.

A five-minute audit

Open the five apps you use most and check three settings in each: what data it collects, whether it keeps a history of your voice, location or searches, and whether you can switch that history off or delete it. Most people find at least one app doing more than they assumed.

Where that leaves you

AI is already in your day, but in a few places, not everywhere. Use the sort question, give your scepticism to the claims that deserve it, and be stingy with personal information in any tool you cannot explain. Start tonight with the five apps.

Sources: Office of the Australian Information Commissioner, Guidance on privacy and the use of commercially available AI products, 2024 (https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products); Australian Government Department of Industry, Science and Resources, Australia’s AI Ethics Principles, 2019 (https://www.industry.gov.au/publications/australias-ai-ethics-principles)