Older people combine two characteristics: they stand to benefit most from a competent voice interface, and they are among those most exposed to the misuse of these same technologies. Addressing the issue honestly means holding both together, without paternalism or alarmism.
We have discussed accessibility and the rise of voice as an interface. Here is a population where these two issues intersect particularly sharply.
What genuinely helps
The voice interface. This is the central point. Someone who cannot master a keyboard, whose eyesight is failing, or whose arthritis makes handling difficult, finds themselves shut out of a growing share of administrative and social life. Being able to speak naturally and be understood despite hesitations removes a real barrier, as we described in our article on voice.
Administrative assistance. Understanding a letter, filling in a form, knowing who to contact. Many cases of unclaimed entitlements stem from the complexity of procedures, not from a lack of information.
Reminders and follow-up. Medication, appointments, errands. Without judgement and without weariness, which matters when the same question keeps coming back.
Companionship, with nuance. This is the delicate point. An available interlocutor can ease real loneliness, and it would be condescending to deny that. But we have explained in our article on loneliness why this can make human contact comparatively more costly. The risk is higher for someone whose contacts are already scarce.
The main risk: voice scams
The most well-documented danger needs to be named clearly. Voice cloning can now reproduce a voice from a short sample, often findable online.
The typical scenario is a call from a relative in distress, demanding an urgent transfer. It already worked before AI. With a recognisable voice, its persuasive power changes in nature.
The countermeasure is simple and effective: agree within the family on a question whose answer only members know, and use it systematically whenever a call demands money urgently. This is more reliable than any technical detection, and it costs nothing to set up.
An AI asserts itself with the same confidence whether it is right or wrong, as we detailed on calibration. For an isolated person, with no one around to cross-check information, this flaw is more dangerous than elsewhere. An erroneous answer about a medication, a right, or a procedure may never be corrected. This is exactly the paradox we noted regarding access to law: those who benefit most from the tool are those least able to verify its outputs.
What works in practice
A few principles drawn from what is done best in this area.
Configure rather than train. Spending an hour properly setting up a device, with voice enabled and unnecessary features removed, is better than lengthy explanations of how to navigate menus.
One use at a time. Introduce one useful function that solves a real problem, rather than a general-purpose assistant that serves no specific purpose.
Keep human contact around it. The tool should make it easier to call family, not replace it. A reminder that suggests calling someone is better than a reminder that fills the time one would have spent calling.
What to take away
There is an injustice in this situation, and it deserves to be named. Those who would gain the most from these tools are those least equipped to protect themselves from their downsides, and those least on the minds of their designers.
The answer is not to keep them away from these tools, which would add one more exclusion to those that already exist. It lies rather in the people around them: someone who configures, who explains once the risk of the cloned-voice call, and who remains reachable. That is hardly technological, and it is probably what protects best.