The delicious irony at the heart of trusting AI

Earlier this week, I set out to read a survey on trust in AI published by Micro Center News. I found the page, loaded it up, and was immediately stopped - not by a lack of interest, but by an AI-powered challenge designed to keep bots out (commonly known as a CAPTCHA - “Completely Automated Public Turing test to tell Computers and Humans Apart”).

As a blind screen reader and keyboard user, I did what I always do: tabbed to the checkbox and pressed Space. And then again. And again. Each time, I was blocked. The system didn’t recognise my interaction as “human enough”.

My understanding is that these challenges have an invisible ‘hot zone’ around the checkbox that looks for human variability as the mouse pointer approaches. But I can’t use a mouse, so I tried the only thing I could, which was adjusting my timing when tabbing to the box and pressing Space to check it - adding the kind of variability that might pass for human behaviour. Still no success. Eventually, after several minutes, I gave up. 

But then I did something else: I asked an AI tool to summarise the page for me. And it did - quickly, accurately, and without friction. In other words, the AI had no trouble accessing the content that I, a  human, was prevented from reaching. I have no idea how this non-human tool bypassed a check specifically designed to keep it out but, as a digitally-excluded, disabled human, I’m very glad that it did.

It’s hard to ignore the irony here. When trying to read an article exploring our trust in AI, an AI-powered system designed to tell humans and machines apart had done exactly that - just in reverse.

Even more striking is the dual role AI played here. It was both the barrier and the solution. On the one hand, it denied my humanity. On the other, it helped me overcome a barrier created without me in mind.

That tension feels important. Because while this experience is, on one level, frustrating (and yes, a little absurd), it also offers a very real glimpse into our evolving relationship with AI. One where trust, capability, and inclusion are not always aligned - and where the outcomes can be contradictory, even within a single interaction.

AI is changing everything - for better and for worse

We are living through a period of profound technological change. AI is no longer a niche capability or a background optimisation, but instead is rapidly becoming embedded in how we work, learn, communicate, and make decisions.

In many cases, this is transformative in positive ways. AI can:

• remove friction from everyday tasks 
• provide instant access to information 
• support creativity and productivity 
• enable more personalised experiences 

For disabled people, and for older people who are disproportionately affected by digital exclusion, the potential is particularly significant.
AI-powered tools can:

• convert text to speech and speech to text in real time 
• simplify complex information 
• assist with memory, organisation, and communication 
• provide new ways of interacting with digital services 

In short, AI has the potential to unlock greater independence, confidence, and participation in work, education, and daily life.
But as my experience shows, the same technologies can also introduce new barriers.

When systems are not designed inclusively, or when assumptions about “normal” human behaviour are baked into them, people can be excluded - sometimes in subtle ways, sometimes completely.
And this is where trust becomes critical.

Because the benefits of AI are not just about what it can do, but about whether people feel able - and safe - to rely on it.

The rise of agentic AI - and why trust matters even more 

Looking ahead, one of the most significant developments in AI is the emergence of agentic systems - AI that doesn’t just respond to prompts, but can take actions on our behalf.
This might include:

• managing schedules and communications 
• completing transactions 
• navigating complex processes 
• making recommendations with increasing autonomy 

For many people, particularly those who face barriers in traditional digital environments, this could be game-changing.

Imagine an AI assistant that can independently navigate inaccessible websites, complete forms, or advocate on your behalf in systems that were not designed with you in mind. That is a powerful prospect that’s becoming a reality.
But it also raises the stakes considerably, because agentic AI requires a different level of trust. It’s not just about whether the information is accurate. You need to be able to trust the system will act in your best interests, respect your preferences, and avoid causing harm.

For disabled users, who may already have experienced exclusion or bias in digital systems, this trust cannot be taken for granted. It must be earned - through transparency, reliability, inclusive design, and clear accountability.
Without that trust, the very people who stand to benefit most from these technologies may be the least able to use them.

So what did the survey say about trust in AI?

In case anyone is left wondering what the survey actually said, we’ll finish off with the key takeaways that the AI was able to pull out for me - even though I wasn’t able to access it myself. Based on over 1000 US participants, the survey found:

1. “Trust but verify” is the dominant pattern - and verification is inconsistent

Most Americans rely on AI regularly, but they do not verify it consistently. Nearly one in five respondents say they always double-check AI output, while 35% only verify if something seems questionable. This suggests that AI is often trusted by default rather than critically assessed.

This creates a risk profile where inaccuracies can slip through unnoticed, especially in routine or low-friction uses.

2. People use AI for medical advice more than they trust it

A striking tension appears around healthcare: only 30% of respondents say they trust AI for medical advice, yet 39% report using it for medical guidance.

This gap highlights behaviour driven by convenience or access rather than confidence - raising concerns about risk, over-reliance, and the need for stronger guidance.

3. Confidence in spotting AI-generated images is inflated

While 57% of respondents said they were confident they could identify AI-generated images, only 55% correctly identified which image was AI-generated when tested.

Additionally, 27% of Americans say they are fooled by AI-generated content on social media at least once a day - underscoring how easily synthetic media blends into everyday digital consumption.

4. AI is trusted more than people in some problem-solving contexts

Then asked who is best at problem-solving, many respondents ranked AI above friends, family, and even therapists.

More than half of respondents also use AI to polish their work, and around one in ten rely on it heavily as part of regular workflows - reinforcing the idea that AI is no longer a background tool, but an active decision-making partner.

Taken together, the survey shows that trust in AI is widespread, situational, and often unearned - particularly given inconsistent verification and overconfidence in recognising AI-generated content.

Building trustworthy, inclusive AI - what comes next 

What strikes me most is how closely these findings mirror my own experience this week.

AI is already embedded in our lives in ways that are both empowering and problematic. It can exclude and include, mislead and assist, frustrate and enable - sometimes all at once.

The question is not whether we will use AI. We already are. The question is whether we can build systems - and experiences - that are worthy of our trust.

That doesn’t happen by accident. It requires deliberate choices - to design inclusively, to test with diverse users, to challenge assumptions about what “normal” behaviour looks like, and to ensure that accessibility is not treated as an afterthought.

For organisations, there is a clear call to action here: make inclusion and accessibility a core part of your AI strategy from the outset, not something you retrofit later.

This is exactly where AbilityNet can help.

With our range of expert, AI-focused consultancy, AbilityNet supports organisations to design and deploy AI-driven tools and services that work for everyone. That means not only meeting legal and ethical obligations, but unlocking the full potential of AI to empower users rather than exclude them.

Because if AI is going to play an increasingly central role in how we live and work, then trust is not a “nice to have”. It’s something we have to actively design for, test for, and be accountable for.

And if we get that right, AI doesn’t just become more usable… it becomes more human.

Let us help you on your AI journey. Please see our range of AI integration services and contact us for more information.