Can distrust in AI disrupt your business?

Are you scared yet, human? 

Arti­fi­cial intel­li­gence (AI) has pro­lif­er­at­ed with trans­for­ma­tive effects in recent years, in sec­tors from autonomous vehi­cles to per­son­alised shop­ping. But the lat­est deploy­ment of AI to gen­er­ate con­tent such as text, images or audio has caused quite a stir. 

Chat­G­PT, a par­tic­u­lar­ly supe­ri­or lan­guage mod­el, even passed the US med­ical spe­cial­i­ty exam. That’s not to say there haven’t been some bloop­ers. The Ope­nAI release has deliv­ered inac­cu­rate infor­ma­tion and even abuse. It also self-warns that it could gen­er­ate bias and harm­ful instruc­tions.

Even before tools such as Chat­G­PT, Bard and Dall‑E 2 attract­ed wide atten­tion, there have been con­cerns that dis­crim­i­na­tion and bias are baked into algo­rithms. Plus, smart tech that seems to pre-empt what you want or make deci­sions based on your online pres­ence is some­what Black Mir­ror-esque. 

This phase of new­er, more acces­si­ble AI could sig­nif­i­cant­ly impact users’ trust in the tech­nol­o­gy. What, then, for busi­ness­es that have been rush­ing to adopt these lat­est forms of AI – should they count on its longevi­ty and if so, is it pos­si­ble to embed AI ethics from the out­set, so that mis­trust doesn’t hurt their rep­u­ta­tion and bot­tom line?

Is AI vital for business?

Despite some opin­ions that gen­er­a­tive AI is a fad, most peo­ple think it’s here to stay. Still, as a Morn­ing Con­sult sur­vey of 10,000 US adults revealed, only 10% of the pub­lic find gen­er­a­tive AI “very trust­wor­thy”. Drilling down fur­ther, that lev­el of trust wavers between demo­graph­ic groups, with younger cohorts, pri­mar­i­ly male, more trust­ful and will­ing to adopt ear­ly than old­er gen­er­a­tions, who are gen­er­al­ly hes­i­tant to pick up new tech­nol­o­gy.

It isn’t only the end users who har­bour doubts, though. High-pro­file gaffes – such as when Google’s Bard cir­cu­lat­ed false facts in search results – have per­haps been a cau­tion­ary tale. Apple Inc also delayed approv­ing updates to its email app with AI-pow­ered lan­guage tools, over con­cerns that it might show chil­dren inap­pro­pri­ate con­tent.

That’s not to say that gen­er­a­tive AI isn’t huge­ly use­ful and it is being applied across busi­ness func­tions, from mar­ket­ing and sales to IT and engi­neer­ing. Its appli­ca­tions range from craft­ing text to cut­ting through dense mate­r­i­al to aid under­stand­ing and answer com­plex ques­tions.

“Com­pa­nies invest a lot in data and tech,” says Karl Weaver, SVP con­sult­ing, EMEA, at Medi­aLink. But he warns: “There’s a gen­er­al accep­tance of ana­lyt­ics now and poten­tial­ly a fear of miss­ing out – busi­ness­es see what the com­pe­ti­tion is doing and think maybe they should do the same. All of that could cause a mis­step and sub­se­quent­ly a trust prob­lem.” 

That’s not to say busi­ness­es should – or even can – avoid the wave. But CEOs and boards should step back and think about why they might be using AI tools, includ­ing these most recent iter­a­tions. If it’s a gen­uine desire to improve cus­tomer expe­ri­ence, for instance, it then should be set up to ensure they’re serv­ing that area.

Practical steps for trustworthy AI

“We are start­ing to devel­op a uni­fied and com­mon under­stand­ing of the big risks,” says Robert Grosvenor, a man­ag­ing direc­tor at Alvarez & Marsal, “but there’s still a long way to go to trans­late high-lev­el, prin­ci­ple-based objec­tives into cod­i­fied require­ments, stan­dards and con­trols.”

The scope and scale of AI’s appli­ca­tion span dif­fer­ent indus­tries and sec­tors and the pos­si­bil­i­ties of harm and degree of risk vary accord­ing­ly; impacts could even dif­fer with­in the same organ­i­sa­tion and be unfore­see­able. As a result, indi­vid­ual busi­ness func­tions could have unique work­flows for their use of AI and the data they need, instead of rely­ing on ana­lyt­ics or com­pli­ance func­tions to dic­tate a cook­ie-cut­ter approach to rules about using AI.

Andrew Strait is asso­ciate direc­tor of emerg­ing tech­nol­o­gy and indus­try prac­tice at the Ada Lovelace Insti­tute, which research­es the impact of data and AI on peo­ple and soci­ety. He says that dis­trust in some AI tech­nolo­gies has meant peo­ple want to see more reg­u­la­tion. Con­sumers can be con­fi­dent that the food they buy in a super­mar­ket is rel­a­tive­ly safe, for exam­ple. But the same lev­el of reg­u­la­to­ry over­sight, and thus con­sumer trust, doesn’t exist for AI – the tech­nol­o­gy has devel­oped too fast for the reg­u­la­tion to keep up.

Strait says that peo­ple want trans­paren­cy around the data prac­tices involved in AI as well as indi­vid­ual pri­va­cy but there’s often the mis­con­cep­tion that telling peo­ple what you are doing is enough for trust to be built. “That lacks a deep under­stand­ing of the con­text in which some­one is expe­ri­enc­ing your prod­uct.”

European Union leads the way in AI standards

What would be good to see, he says, is people’s par­tic­i­pa­tion in the gov­er­nance of AI. 

Data coop­er­a­tives could be the means to that end. A rep­re­sen­ta­tive col­lec­tive of peo­ple in a data set decides who access­es the data. In the Span­ish health­care space, for instance, the coop­er­a­tive Salus Coop gives cit­i­zens con­trol over their data for research pur­pos­es.

Despite the gen­er­al reg­u­la­to­ry lag, the EU is tack­ling the prob­lem. The so-called AI Act (Reg­u­la­tion Lay­ing Down Har­monised Rules on Arti­fi­cial Intel­li­gence) is under dis­cus­sion and looks to address eth­i­cal dilem­mas and safe­guards, such as by assign­ing risk lev­els to AI’s var­i­ous deploy­ments, while enabling AI’s ben­e­fits.

Gen­er­a­tive AI tools for use in sen­si­tive areas such as recruit­ment would be classed as a “high risk” des­ig­na­tion. This would enforce “con­for­mi­ty assess­ments” to meet cer­tain stan­dards, which would reas­sure the pub­lic. The Euro­pean bloc’s yard­stick could well be sought fur­ther afield and influ­ence the reg­u­la­to­ry envi­ron­ment of the UK or beyond.Human appre­hen­sion about AI, or a hes­i­tan­cy to ful­ly trust it, per­sists. But the real­i­ty is that while we attribute human-like expec­ta­tions to AI, it is just machin­ery. As actu­al humans, we can still check facts from fic­tion and set expec­ta­tions around accept­able uses of AI and busi­ness­es can lead that charge.

Are you scared yet, human? 

Artificial intelligence (AI) has proliferated with transformative effects in recent years, in sectors from autonomous vehicles to personalised shopping. But the latest deployment of AI to generate content such as text, images or audio has caused quite a stir. 

ChatGPT, a particularly superior language model, even passed the US medical speciality exam. That’s not to say there haven’t been some bloopers. The OpenAI release has delivered inaccurate information and even abuse. It also self-warns that it could generate bias and harmful instructions.