Some time ago, I wrote a col­umn called Ter­mi­nat­ing Bar­ri­ers. You should read it if you haven’t.

In it, I talked about the ways blind and vi­su­al­ly im­paired (BVI) peo­ple were us­ing Large Lan­guage Mod­els (LLMs/AI) as tools of em­pow­er­ment.

In gen­er­al, the point re­mains the same. We’re still us­ing it in cre­ative ways to nav­i­gate dai­ly life. Be it find­ing food in the fridge or de­scrib­ing friends in their Lilo & Stitch py­ja­mas.

The on­ly change since is con­stant im­prove­ment.

In that col­umn, I said there were down­sides of AI. You’re prob­a­bly think­ing of the ones we al­ways hear about:

“The ro­bots will take our jobs”.

“The ro­bots are us­ing up all our wa­ter”.

“What if the ro­bots think hu­man­i­ty is a threat to the plan­et and must be liq­ui­dat­ed?”

All of these are im­por­tant, and, peo­ple smarter than me are try­ing to deal with these is­sues. But as usu­al with the BVI com­mu­ni­ty, this tech is game chang­ing, though con­verse­ly, has a unique­ly com­pli­cat­ed im­pact on us.

These days, big tech cor­po­ra­tions wield more in­flu­ence over our lives. When Meta or Google has an out­age, every­body feels it. No What­sApp. No YouTube. No smug­ly telling some­one on Twit­ter (X) to “Just Google it, Bro”.

Many of our AI fea­tures come to us from Big Tech such as Google Gem­i­ni Live. So, as you can imag­ine, these out­ages im­pact us more dra­mat­i­cal­ly than chat­ting on What­sApp or watch­ing fun­ny memes on Face­book.

And yes, we do watch fun­ny memes on Face­book.

Dur­ing the out­age, it’s back to the same old, same old. Back to need­ing some­one’s help to read a let­ter or med­i­cine in­struc­tions. Back to fum­bling around to find some­thing that’s been mis­placed. Back to be­ing de­pen­dent.

That’s the head­line con­cern for BVI peo­ple when we talk about these im­por­tant tools be­ing in the hands of just a few cor­po­ra­tions, but what about the fine print?

Do you know that if I gave some AI de­scrip­tion tools my pic­ture and asked it if the per­son in the im­age is a young man or young woman, some­times it won’t tell me?

That’s right. The sec­ond you looked at the pho­to you al­ready knew the an­swer. Mean­while, a blind per­son is left guess­ing.

This isn’t a lim­i­ta­tion. Far from it. The AI won’t tell you that Jaidon Vin­cent is a young man not be­cause it can’t. In­stead, it won’t tell you be­cause in an of­fice some­where, a group of cor­po­rate ex­ec­u­tives de­cid­ed it was bet­ter to de­ny blind peo­ple such ba­sic in­for­ma­tion. That it was safer to de­cide what’s ap­pro­pri­ate and in­ap­pro­pri­ate for blind peo­ple to have de­scribed to them.

My favourite ex­am­ple of this is that time some im­age de­scrip­tion tools re­fused to process any pho­to with a per­son. Ap­par­ent­ly, there was a glitch flag­ging hu­mans as in­ap­pro­pri­ate adult con­tent.

Imag­ine if a soft­ware up­date for your eyes re­leased to­mor­row la­belling walls as in­ap­pro­pri­ate.

That would sure­ly go well.

You may won­der, if it doesn’t gate­keep or hal­lu­ci­nate 75 per cent of the time, what’s the is­sue? The is­sue is if 1 in 20 bot­tles of a brand of wa­ter was poi­son, would you buy it?

If you’re not will­ing to ac­cept a 1 in 20 risk, why should we set­tle for 1 in 4?

This isn’t a mi­nor in­con­ve­nience nor hy­per­bole. This is a case of hav­ing our world, our re­al­i­ty, cu­rat­ed in a way that we don’t want, we don’t need, or may even be harm­ful to us.

Across his­to­ry, there’ve been var­i­ous re­source booms. Oil, sug­ar, even bird poop. The lat­est seems to be user da­ta.

Just like the booms be­fore it, there are ques­tions about how our da­ta is har­vest­ed and stored.

You may be see­ing where I’m go­ing.

Blind peo­ple use AI to make our lives eas­i­er, no dif­fer­ent to the rest of so­ci­ety. The dif­fer­ence is we use it more in­ti­mate­ly than sight­ed folks.

These com­pa­nies have footage of our bed­rooms. They have an idea of the soft­ware we use and how we use it and, they see the in­ti­mate, crit­i­cal parts of life that you wouldn’t want an­oth­er per­son to see.

Sight­ed peo­ple will nev­er have these con­cerns be­cause they don’t need AI in the way we do. For most peo­ple, it’s just a cu­rios­i­ty. The most per­son­al in­for­ma­tion they may give is strange role play or ask­ing a sil­ly ques­tion.

Sight­ed peo­ple have no need for an AI to tell them, and con­se­quent­ly, for the par­ent com­pa­ny to know about, the heart-warm­ing smile of their ba­by in the bath­tub.

They just watch and melt.

Mean­while, we point the cam­era, wait for a de­scrip­tion, and hope it tells us the cor­rect thing.

The cen­tral is­sues defin­ing the re­la­tion­ship be­tween the blind and AI is the in­fra­struc­ture they run on be­ing vul­ner­a­ble and prone to fail­ure, and, AI be­ing cre­at­ed by and main­tained by, busi­ness­es rather than hu­man­i­tar­i­an or­gan­i­sa­tions. Even where AI is lever­aged by a blind peo­ple first or­gan­i­sa­tion like Be My Eyes, the un­der­ly­ing tech is con­trolled by a busi­ness.

The cal­cu­la­tions are dif­fer­ent for them.

Cal­cu­la­tions of li­a­bil­i­ty and risk. Of in­fra­struc­ture costs and a po­ten­tial­ly neg­a­tive news­pa­per col­umn. Not pri­va­cy, nor en­sur­ing BVI peo­ple live their best and hap­pi­est lives our way.

These neg­a­tive columns are need­ed though. Not to com­plain, not to make you feel sad about your priv­i­lege nor self-con­scious about un­usu­al con­ver­sa­tions with AI.

These con­ver­sa­tions are im­por­tant be­cause this isn’t a 20-year fore­cast which gets dras­ti­cal­ly re­vised every 47 days.

These are re­al is­sues fac­ing re­al peo­ple.

Now.

This col­umn is sup­plied in con­junc­tion with the T&T Blind Wel­fare As­so­ci­a­tion.

Head­quar­ters: 118 Duke Street, Port of Spain, Trinidad

Email: ttb­wa1914@gmail.com

Phone: (868) 624-4675

What­sApp: (868) 395-3086