Every second brochure that crosses my desk now promises “artificial intelligence,” but here’s what that actually means in a hearing aid: machine learning algorithms that automatically adjust amplification based on your acoustic environment and gradually learn which settings you prefer. It’s genuinely useful technology—less manual fiddling, faster adaptation to new cafés or meeting rooms—but it’s not HAL 9000, and the marketing departments would very much like you to believe it’s more revolutionary than it is.
After fitting several hundred AI-enabled devices over the past few years, I can tell you exactly what these systems do well, where they fall short, and which features are worth paying attention to when you’re comparing devices. Because the gap between the glossy brochure and the follow-up appointment six weeks later is where most of the confusion lives.
What “AI” Actually Means in Your Hearing Aid
Let’s start with what we’re really talking about. The AI in modern hearing aids is narrow machine learning—algorithms trained on massive datasets of acoustic environments that can classify the sound around you and adjust processing in real time. When you walk from a quiet lounge room into a noisy restaurant, the device detects the change (analysing things like signal-to-noise ratio, temporal patterns, frequency distribution) and shifts to a different processing profile within seconds. It’s doing what you used to do manually by pressing a button to change programs, except it’s faster and often more accurate.
The second component is learning your preferences. Some systems—Widex’s SoundSense Learn is the clearest example—present you with pairs of sound settings via a smartphone app and ask which you prefer. Over time, the algorithm builds a profile of how you like speech clarity weighted against background softness, or how much high-frequency gain you’ll tolerate before it sounds tinny. These preferences are fed back into the automatic adjustments, so the aid becomes progressively more aligned with your actual listening habits rather than just the prescription calculated from your audiogram.
What this is *not* is general artificial intelligence. Your hearing aid isn’t thinking, reasoning, or understanding speech. It’s not going to have a conversation with you about why it made a particular adjustment. It’s pattern recognition—sophisticated pattern recognition, trained on thousands of hours of labelled audio, but pattern recognition nonetheless.
Current Capabilities: What These Devices Can Actually Do
The technology has genuinely improved over the past four or five years, and I say that as someone who’s sceptical of most “breakthrough” claims in this industry. Modern AI systems can identify twelve to fifteen distinct acoustic environments—things like “quiet conversation,” “restaurant,” “traffic noise,” “music,” “large meeting,” “outdoor”—and they’re right most of the time. When I verify this with real-ear measurement (placing a probe microphone in the ear canal while the aid is operating), the frequency response genuinely shifts in ways that align with what we’d want for that environment.
Conversation enhancement has become noticeably better. Devices like the Oticon Intent and Starkey Genesis AI use sensor data—some track head movement, others incorporate accelerometers to detect whether you’re walking or sitting still—to work out which direction you’re focusing on. If you’re in a noisy environment and you turn your head toward someone speaking, the directional microphones sharpen their focus in that direction within a second or two. It’s not perfect, especially in really dreadful acoustics like venues with hard floors and glass everywhere, but it’s a measurable improvement over fixed directional patterns.
Noise classification is where things get interesting from a clinical perspective. The aids can differentiate between steady-state noise (aircon hum, traffic drone) and modulated noise (clinking plates, background babble), and they handle each type differently. Steady noise gets more aggressive reduction; modulated noise gets lighter treatment because too much suppression makes everything sound hollow and can actually cut into speech cues. When I run speech-in-noise testing—usually the QuickSIN or the adaptive BKB sentences—patients fitted with good AI systems tend to score one to two dB better than with older premium devices. That’s the difference between catching 70% versus 80% of words in a typical noisy restaurant, which is a meaningful real-world change.
What the Marketing Materials Won’t Tell You
Here’s what I see in follow-up appointments. The automatic scene detection works well in *distinct* environments, but in ambiguous situations—say, a café that’s moderately busy but not packed—the device can hunt between settings. You’ll hear it shift from “restaurant” to “speech in noise” to “quiet” and back again over the course of five minutes. Most patients don’t find this particularly bothersome once they know it’s happening, but it’s not the seamless experience the brochures suggest.
The learning algorithms are only as good as the feedback you give them. If you never use the app to rate sound quality or adjust settings, the system has nothing to learn from. I’ve had patients return after three months saying “it hasn’t learned anything” when they’ve never opened the app. The machine learning isn’t telepathic—it needs data.
Battery life takes a hit with AI processing. The more sensors, the more active the machine learning, the faster the battery drains. Rechargeable devices that used to last 22-24 hours might now give you 18-20 with all the AI features active. It’s not usually a problem—most people aren’t wearing aids for twenty straight hours—but if you’re a shift worker or travelling across time zones, it’s something to plan for.
The Brands Actually Using AI (And How They Differ)
Starkey’s Edge AI platform leans heavily on environment classification and uses quite a bit of sensor data (they were early adopters of accelerometers and fall detection in hearing aids). Their system is particularly good at detecting and enhancing speech in complex environments—I’ve had several patients with moderate-to-severe high-frequency loss report genuine improvement in group conversations after switching to Genesis AI devices. The trade-off is that Starkey’s approach can be aggressive, sometimes at the expense of music fidelity. If you’re someone who listens to a lot of acoustic music or plays an instrument, it’s worth spending extra time in the demo period specifically testing that.
Oticon’s Deep Neural Network (used in the Intent and Real ranges) was trained on twelve million sound scenes, which sounds like marketing hyperbole but actually translates to slightly better performance in unusual acoustic environments—places the algorithm hasn’t specifically been trained on. Their philosophy is to preserve more of the sound scene rather than aggressively isolate speech, based on the theory that the brain benefits from access to fuller information. In practice, this means Oticon devices tend to sound more “open” and natural, but some patients find they have to work harder to focus in very noisy settings. It’s a trade-off, and which approach suits you is genuinely individual.
Widex’s SoundSense Learn is the most transparent of the learning systems because you’re directly training it via A/B comparisons. You get two sound settings, you pick the one you prefer, and over twenty or thirty comparisons the algorithm hones in on your preferences. I find this works particularly well for patients who are very clear about what bothers them—”I hate when it sounds metallic,” “I need more bass”—because they’re actively shaping the output. The downside is it requires engagement. If you want a device you can just put in and forget, the Widex learning system might feel like homework.
What to Actually Test When You’re Trying These Devices
When you’re demoing AI hearing aids—and you absolutely should demo them before committing, given the cost (typically $4,500 to $7,500 per pair for premium AI-enabled devices in Australia)—here’s what to pay attention to. First, go somewhere with genuinely challenging acoustics. Not the quiet clinic room, but an actual café at lunchtime or a busy shopping centre. Wear the aids for at least thirty minutes in that environment. The AI needs time to settle and adapt, and you need time to notice whether the adjustments are helping or just creating distraction.
Second, test transitions. Walk from inside to outside, from a quiet room into a noisy one. Does the aid adapt smoothly, or does it overreact and make everything sound muffled for ten seconds? The quality of these transitions varies significantly between brands and even between models within the same brand.
Third, if the device has a learning feature, actually use it during the trial period. Rate sounds, make adjustments, give the algorithm something to work with. Then assess whether the experience improves over the second and third week compared to the first few days. If you’re not noticing any adaptation, that tells you something about either the algorithm or how well it aligns with your preferences.
The Bottom Line From the Clinic Room
AI in hearing aids is real technology that provides measurable benefits—primarily less need for manual program switching and somewhat better performance in complex listening environments. It’s not magic, and it’s not going to give someone with a severe high-frequency loss perfect hearing in a crowded pub. The improvements are incremental but genuine, typically in the range of 10-20% better speech understanding in difficult situations compared to previous-generation premium devices.
Whether those improvements are worth the price premium (AI-enabled devices typically cost $1,000 to $2,000 more per pair than equivalent non-AI models) depends on your lifestyle and listening demands. If you spend most of your time in quiet or moderately noisy environments, you might not notice enough difference to justify the cost. If you’re regularly in challenging acoustics—lots of meetings, social gatherings, complex work environments—the adaptive features can meaningfully reduce listening effort.
We keep demo units of the major AI platforms at The Audiology Place precisely because this technology is difficult to evaluate from spec sheets. You need to hear it in your life, not in a brochure or a perfectly controlled listening test. The AI features that matter are the ones that work for *your* specific hearing loss, in *your* actual acoustic environments, with *your* tolerance for tinkering versus automation. Everything else is marketing.
Always read the instructions for use. Follow the instructions for use.
