Could AI help us create imagination machines?

Human cre­ativ­i­ty is the elixir that’s pow­ered civil­i­sa­tions down the ages. It’s brought us untold break­throughs in all sec­tors of the econ­o­my, from agri­cul­ture to health­care, ener­gy to mobil­i­ty. Our imag­i­na­tion con­tin­ues to be our sav­iour as the world’s age­ing pop­u­la­tion faces tough socioe­co­nom­ic and envi­ron­men­tal chal­lenges. 

Imag­i­na­tion entails cre­at­ing men­tal mod­els of things that don’t yet exist. This kind of inno­va­tion brought us the print­ing press, the steam engine, the light bulb, the tele­phone, the aero­plane, the TV and the PC. All of these required improb­a­ble acts of human inven­tion. Yet many busi­ness­es are fail­ing to har­ness their employ­ees’ cre­ativ­i­ty, even though they say it’s a vital trait for suc­cess.

Mar­tin Reeves, co-author of a recent­ly pub­lished book called The Imag­i­na­tion Machine: how to spark new ideas and cre­ate your company’s future, is chair of the BCG Hen­der­son Insti­tute, a think-tank cre­at­ed by the Boston Con­sult­ing Group. He believes that “we need to focus on imag­i­na­tion because a com­pet­i­tive advan­tage doesn’t last very long any­more. If you were the leader of your indus­try in the 1980s, you could expect to be at the top for at least 10 years. Now that peri­od has come down to one to two years.”

He con­tin­ues: “This means that com­pa­nies can’t just focus on opti­mis­ing yesterday’s busi­ness mod­el; they need to cre­ate new ones. We must gen­er­ate growth through our cre­ativ­i­ty.”

Com­put­ers still strug­gle to emu­late this unique­ly human process. Even the most high-pow­ered machines can­not imag­ine a prod­uct or ser­vice that’s nev­er been seen before. They can’t con­jure up the new iPhone or Tes­la of tomor­row. (Nei­ther can many humans, for that mat­ter.) But the tide is turn­ing. 

Humans can dis­as­sem­ble and recom­bine learnt knowl­edge to con­ceive of new images – think of a red boat, visu­alise a blue car and then imag­ine a blue boat, for instance. A team of AI experts from the Uni­ver­si­ty of South­ern Cal­i­for­nia (USC) is research­ing how to emu­late the process arti­fi­cial­ly. This involves a con­cept called dis­en­tan­gle­ment. Humans break down the things they learn and visu­alise them as colours, shapes and types. We are then able to recom­bine these attrib­ut­es to form nov­el images. The researchers have repli­cat­ed this using neur­al net­works. 

“The new approach tru­ly unleash­es a new sense of imag­i­na­tion in AI sys­tems, bring­ing them clos­er to a human under­stand­ing of the world,” says researcher Yun­hao Ge, who has been lead­ing the project at USC.  

Dis­en­tan­gle­ment isn’t a new con­cept. Indeed, it’s being used to cre­ate so-called deep­fake con­tent, typ­i­cal­ly by recom­bin­ing sep­a­rate images of well-known peo­ple. AI has also been used in this way to invent a new sport known as speedgate. Data from 400 pop­u­lar sports was fed into a neur­al net­work. Out popped many sug­ges­tions, but the win­ner involved merg­ing ele­ments of foot­ball, cro­quet and rug­by on a field with six-play­er teams, a ball and three gates.

The USC sys­tem takes a whole group of sam­ple images or data, rather than one at a time, as tra­di­tion­al algo­rithms have done. The sys­tem looks at sim­i­lar­i­ties between those images, which it then recom­bines using all the data to make some­thing new. The process is known as con­trol­lable nov­el image syn­the­sis. 

What’s new is that the sys­tem can han­dle near­ly any kind of data. It may even­tu­al­ly help in the dis­cov­ery of nov­el drugs by com­bin­ing learnt func­tions from exist­ing med­i­cines to form inno­v­a­tive treat­ments. The tech­nol­o­gy could dis­en­tan­gle race and gen­der bias to make fair­er algo­rithms or cre­ate safer dri­ver­less cars by sim­u­lat­ing count­less crash­es. The method may even be able to cre­ate new data sets by imag­i­na­tion.

“In the future, AI could be effec­tive­ly bound­less,” pre­dicts Michael Con­way, leader of the AI and ana­lyt­ics prac­tice at IBM. “It will be able to see con­nec­tions between things that humans can’t. A more advanced ver­sion will be able to look beyond the para­me­ters we have set and offer valu­able insights that humans wouldn’t have even imag­ined.”   

If a more imag­i­na­tive organ­i­sa­tion is a clear goal for busi­ness lead­ers, the future is like­ly to involve tech­nol­o­gy, such as the sys­tem cre­at­ed at USC, that could in time com­ple­ment more cre­ative think­ing by employ­ees. After all, when the algo­rithm was trained to cre­ate speedgate, a human still had to choose the most real­is­tic sport to select from the count­less absurd options that the sys­tem sug­gest­ed. Aug­ment­ed imag­i­na­tion is there­fore like­ly to be the next step. The term that Reeves uses to describe a busi­ness that can gain such capa­bil­i­ties is “a bion­ic organ­i­sa­tion”. 

“What will work best is AI oper­at­ing along­side humans, rather than inde­pen­dent­ly,” says Melis­sa Ter­ras, pro­fes­sor of dig­i­tal cul­tur­al her­itage and a fel­low of the Alan Tur­ing Insti­tute. “The obvi­ous way for­ward is to have AI process vast amounts of infor­ma­tion and pro­pose new solu­tions and to have a human work­ing in tan­dem, fine-tun­ing the results. We’ll even­tu­al­ly get to know the quirks of these sys­tems and be able to play them like musi­cal instru­ments, as well as choose which ideas need devel­op­ment. This is the ‘imag­i­na­tion machine’ – an act of co-cre­ation, with com­pu­ta­tion­al pow­er and intel­li­gence com­ple­ment­ing our own.”

Humans are rel­a­tive­ly good at coun­ter­fac­tu­al think­ing. This is the abil­i­ty to imag­ine some­thing that doesn’t cur­rent­ly exist but could in the near future. This is what sets us apart from any cur­rent form of machine learn­ing. Yet cor­po­ra­tions aren’t nor­mal­ly focused on this type of activ­i­ty. Instead, they tend to be nose down in quar­ter­ly earn­ings reports, deal­ing with the here and now or fac­ing up to the imme­di­ate past. The imag­i­na­tive spir­it is rarely used to help steer busi­ness­es with a firm eye on the future, but this could change. 

“I am very bull­ish about the bion­ic organ­i­sa­tion – a more effec­tive busi­ness that taps into both human and machine cog­ni­tion,” Reeves says. “The ‘imag­i­na­tion machine’ isn’t just the sil­i­con; it has a human ele­ment. If you are think­ing about growth right now, you are think­ing about imag­i­na­tion.”  

He con­tin­ues: “In 10 years’ time, when some­one says ‘organ­i­sa­tion’, they’ll be refer­ring to a syn­er­gis­tic com­bi­na­tion of algo­rith­mic and human think­ing that gets the job done bet­ter. This rais­es all sorts of ques­tions. For instance, what tech­nol­o­gy will be deployed? How do we use it? How do we match the band­width of machines and humans? How do you ensure that the whole organ­i­sa­tion is serv­ing human ends and is eth­i­cal? What do humans do and what do machines do?”

The cur­rent era of AI is all about using algo­rithms, neur­al net­works and deep learn­ing to auto­mate low­er-val­ue, high­er-vol­ume tasks. The scale of analy­sis that com­put­ers can under­take puts machines in a league of their own when it comes to these kinds of jobs. One school of thought is that apply­ing AI to more of this work could give humans the free­dom to be more cre­ative. But not every­one agrees.

“There is a con­cern that an AI-gen­er­at­ed inno­va­tion sec­tor may be repet­i­tive­ly iter­at­ing on the world that already exists, rather than tran­scend­ing it,” says Ter­ras, who is also a co-direc­tor of the Uni­ver­si­ty of Edinburgh’s R&D pro­gramme in cre­ative infor­mat­ics. “Why not let machines do what they are good at – syn­the­sis­ing and pro­cess­ing vast amounts of infor­ma­tion – while humans deal with com­plex­i­ty and nuance to build prod­ucts and ser­vices that will suc­ceed for human­i­ty?”

We still have some way to go before AI mus­cles in on human-led imag­i­na­tive tasks and enters the main­stream in cre­ative think­ing. But, if that time comes, its poten­tial to ben­e­fit soci­ety could mark a turn­ing point. 

There will be less talk about automa­tion steal­ing people’s jobs or calls for a robot tax when the tech­nol­o­gy can cre­ate the next pan­dem­ic-beat­ing super­drug, say, or an afford­able means of cap­tur­ing and stor­ing atmos­pher­ic CO2. Then AI’s will be firm­ly cement­ed in soci­ety. It will also take human inge­nu­ity to get us there.  

“Will we be able to use organ­i­sa­tion­al sci­ence, brain sci­ence, social sci­ence and dig­i­tal sci­ence to do a bet­ter job? The answer is that we already can,” Reeves says. “The focus of imag­i­na­tion will be the rein­ven­tion of busi­ness.”