The future of brain-machine synchronisation

The remark­able evol­u­tion of Arti­fi­cial Intel­li­gence (AI) sys­tems rep­res­ents a paradigm shift in the rela­tion­ship between humans and machines. This trans­form­a­tion is evid­ent in the seam­less inter­ac­tions facil­it­ated by these advanced sys­tems, where adapt­ab­il­ity emerges as a defin­ing char­ac­ter­ist­ic, res­on­at­ing with the fun­da­ment­al human capa­city to learn from exper­i­ence and pre­dict behaviour.

One facet of AI that aligns closely with human cog­nit­ive pro­cesses is Rein­force­ment Learn­ing (RL). RL mim­ics the human learn­ing paradigm by allow­ing AI sys­tems to learn through inter­ac­tion with an envir­on­ment, receiv­ing feed­back in the form of rewards or pen­al­ties. By con­trast, Large Lan­guage Mod­els (LLMs) play a cru­cial role in pat­tern recog­ni­tion, cap­tur­ing the intric­ate nuances of human lan­guage and beha­viour. These mod­els, such as Chat­G­PT and BERT, excel in under­stand­ing con­tex­tu­al inform­a­tion, grasp­ing the sub­tleties of lan­guage, and pre­dict­ing user intent. Lever­aging vast data­sets, LLMs acquire a com­pre­hens­ive under­stand­ing of lin­guist­ic pat­terns, enabling them to gen­er­ate human-like responses and adapt to some of the user beha­viour, some­times with remark­able accuracy.

The syn­ergy between RL and LLMs cre­ates a power­ful pre­dict­or of human beha­viour. RL con­trib­utes the abil­ity to learn from inter­ac­tions and adapt, while LLMs enhance the pre­dic­tion cap­ab­il­it­ies through pat­tern recog­ni­tion. AI sys­tems based on RL can thus dis­play a form of beha­vi­our­al syn­chrony. At its core, RL enables AI sys­tems to learn optim­al sequences of actions in inter­act­ive envir­on­ments to achieve a policy. Ana­log­ous to a child touch­ing a hot sur­face and learn­ing to avoid it, these AI sys­tems adapt based on the pos­it­ive or neg­at­ive feed­back they receive.

AI agents using deep rein­force­ment learn­ing, such as Google Deep­Mind’s AlphaZero, learn and improve by play­ing mil­lions of games against them­selves, thereby refin­ing their strategies over time. This self-improve­ment pro­cess in AI involves an agent iter­at­ively learn­ing from its own actions and out­comes. Sim­il­arly, in human inter­ac­tions, brain syn­chrony occurs when indi­vidu­als engage in cooper­at­ive tasks, lead­ing to aligned pat­terns of brain activ­ity that facil­it­ate shared under­stand­ing and col­lab­or­a­tion. Unlike AI, humans achieve this syn­chrony through inter­ac­tion with oth­ers rather than themselves.

What’s more, AI sys­tems can also learn from inter­ac­tions with humans. Just as human brain syn­chrony enhances cooper­a­tion and under­stand­ing, AI sys­tems can improve and align their responses through extens­ive iter­at­ive learn­ing from human inter­ac­tions. While AI sys­tems do not lit­er­ally share know­ledge as human brains do, they become repos­it­or­ies of data inher­ited from these inter­ac­tions, which cor­res­ponds to a form of know­ledge. This pro­cess of learn­ing from vast data­sets, includ­ing human inter­ac­tions, can be seen as a form of ‘col­lect­ive memory’. This ana­logy high­lights the poten­tial for AI sys­tems to evolve while being influ­enced by humans, while also influ­en­cing humans through their use, indic­at­ing a form of ‘com­pu­ta­tion­al syn­chrony’ that could be seen as an ana­logue to human brain synchrony.

In addi­tion, AI sys­tems enabled with social cue recog­ni­tion are being designed to detect and respond to human emo­tions. These ‘Affect­ive Com­put­ing’ sys­tems, as coined by Ros­alind Picard in 1995¹, can inter­pret human facial expres­sions, voice mod­u­la­tions, and even text to gauge emo­tions and then respond accord­ingly. An AI assist­ant that can detect user frus­tra­tion in real-time and adjust its responses or assist­ance strategy is a rudi­ment­ary form of beha­vi­our­al syn­chron­isa­tion based on imme­di­ate feedback.

For instance, affect­ive com­put­ing encom­passes tech­no­lo­gies like emo­tion recog­ni­tion soft­ware that ana­lyses facial expres­sions and voice tone to determ­ine a person’s emo­tion­al state. Real-time sen­ti­ment ana­lys­is in text and voice allows AI to adjust its inter­ac­tions to be more empath­et­ic and effect­ive. This cap­ab­il­ity is increas­ingly used in cus­tom­er ser­vice chat­bots and vir­tu­al assist­ants to improve user exper­i­ence by mak­ing inter­ac­tions feel more nat­ur­al and responsive.

Just as humans adjust their beha­viour in response to social cues, adapt­ive AI sys­tems modi­fy their actions based on user input, poten­tially lead­ing to a form of ‘syn­chron­isa­tion’ over time. Assess­ing the social com­pet­ence of such an AI sys­tem could be done by adapt­ing tools like the Social Respons­ive­ness Scale (SRS)—a well-val­id­ated psy­chi­at­ric instru­ment that meas­ures how adept an indi­vidu­al is at modi­fy­ing their beha­viour to fit the beha­viour and dis­pos­i­tion of a social part­ner, a proxy for ‘the­ory of mind,’ which refers to the abil­ity to attrib­ute men­tal states—such as beliefs, intents, desires, emo­tions, and knowledge—to one­self and to others.

Brain-Com­puter Inter­faces (BCIs) have ushered in a trans­form­at­ive era in which thoughts can be trans­lated into digit­al com­mands and human com­mu­nic­a­tion. Com­pan­ies like Neur­alink are mak­ing strides devel­op­ing inter­faces that enable para­lysed indi­vidu­als to con­trol devices dir­ectly with their thoughts. Con­nect­ing dir­ect record­ings of brain activ­ity with AI sys­tems, research­ers enabled an indi­vidu­al to speak at nor­mal con­ver­sa­tion­al speed after being mute for more than a dec­ade fol­low­ing a stroke. AI sys­tems can also be used to decode not only what an indi­vidu­al is read­ing but what they are think­ing based on non-invas­ive meas­ures of brain activ­ity using func­tion­al MRI.

Based on these advances, it’s not far-fetched to ima­gine a future scen­ario in which a pro­fes­sion­al uses a non-invas­ive BCI (e.g., wear­able brain­wave mon­it­ors such as Cog­wear, Emotiv, or Muse) to com­mu­nic­ate with AI design soft­ware. The soft­ware, recog­nising the designer’s neur­al pat­terns asso­ci­ated with cre­ativ­ity or dis­sat­is­fac­tion, could instant­an­eously adjust its design pro­pos­als, achiev­ing a level of syn­chrony pre­vi­ously thought to be the realm of sci­ence fic­tion. This tech­no­lo­gic­al fron­ti­er holds the prom­ise of a dis­tinct­ive form of syn­chrony, where the inter­play between the human brain and AI tran­scends mere com­mand inter­pret­a­tion, open­ing up a future in which AI res­on­ates with human thoughts and emotions.

Cru­cially, the res­on­ance envi­sioned here tran­scends the beha­vi­our­al domain to encom­pass com­mu­nic­a­tion as well. As BCIs evolve, the poten­tial for out­ward expres­sions becomes pivotal. Bey­ond mere com­mand exe­cu­tion, the integ­ra­tion of facial cues, tone of voice, and oth­er non-verbal cues into AI’s responses amp­li­fies the chan­nels for res­on­ance. This expan­sion into mul­timod­al com­mu­nic­a­tion may enrich syn­chrony by cap­tur­ing ele­ments from the hol­ist­ic nature of human expres­sion, cre­at­ing a more immers­ive and nat­ur­al interaction.

How­ever, the concept of res­on­ance also presents the chal­lenge of nav­ig­at­ing the uncanny val­ley, a phe­nomen­on where humanoid entit­ies that closely resemble humans pro­voke dis­com­fort. Strik­ing the right bal­ance is para­mount, ensur­ing the AI’s respons­ive­ness aligns authen­tic­ally with human expres­sions, without enter­ing the dis­com­fit­ing realm of the uncanny val­ley. The poten­tial of BCIs to foster syn­chrony between the human brain and AI intro­duces prom­ising yet chal­len­ging pro­spects for human-com­puter collaboration.

Neur­os­cience not only illu­min­ates the basis of bio­lo­gic­al intel­li­gence but may also guide devel­op­ment of arti­fi­cial intel­li­gence². Con­sid­er­ing evol­u­tion­ary con­straints like space and com­mu­nic­a­tion effi­ciency, which have shaped the emer­gence of effi­cient sys­tems in nature, prompts explor­a­tion of embed­ding sim­il­ar con­straints in AI sys­tems, envi­sion­ing organ­ic­ally evolving arti­fi­cial envir­on­ments optim­ised for effi­ciency and envir­on­ment­al sus­tain­ab­il­ity, the focus of research in so-called “neur­omorph­ic computing.” 

For example, oscil­lat­ory neur­al activ­ity appears to boost com­mu­nic­a­tion between dis­tant brain areas. The brain employs a theta-gamma rhythm to pack­age and trans­mit inform­a­tion, sim­il­ar to a postal ser­vice, thereby enhan­cing effi­cient data trans­mis­sion and retriev­al³. This inter­play has been likened to an advanced data trans­mis­sion sys­tem, where low-fre­quency alpha and beta brain waves sup­press neur­al activ­ity asso­ci­ated with pre­dict­able stim­uli, allow­ing neur­ons in sens­ory regions to high­light unex­pec­ted stim­uli via high­er-fre­quency gamma waves. Bas­tos et al.⁴ found that inhib­it­ory pre­dic­tions car­ried by alpha/beta waves typ­ic­ally flow back­ward through deep­er cor­tic­al lay­ers, while excit­at­ory gamma waves con­vey­ing inform­a­tion about nov­el stim­uli propag­ate for­ward through super­fi­cial layers.

Recent AI exper­i­ments, par­tic­u­larly those involving OpenAI’s GPT‑4, unveil intriguing par­al­lels with evol­u­tion­ary learning.

In the mam­mali­an brain, sharp wave ripples (SPW-Rs) exert wide­spread excit­at­ory influ­ence through­out the cor­tex and mul­tiple sub­cor­tic­al nuc­lei⁵. With­in these SPW-Rs, neur­on­al spik­ing is metic­u­lously orches­trated both tem­por­ally and spa­tially by interneur­ons, facil­it­at­ing the con­densed react­iv­a­tion of seg­ments from wak­ing neur­on­al sequences⁶. This orches­trated activ­ity aids in the trans­mis­sion of com­pressed hip­po­cam­pal rep­res­ent­a­tions to dis­trib­uted cir­cuits, thereby rein­for­cing the pro­cess of memory con­sol­id­a­tion⁷.

Recent AI exper­i­ments, par­tic­u­larly those involving OpenAI’s GPT‑4, unveil intriguing par­al­lels with evol­u­tion­ary learn­ing. Unlike tra­di­tion­al task-ori­ented train­ing, GPT‑4 learns from extens­ive data­sets, refin­ing its responses based on the accu­mu­lated ‘exper­i­ences’ – fur­ther­more pat­tern recog­ni­tion by GPTs par­al­lels pat­tern recog­ni­tion by lay­ers of neur­ons in the brain. This approach mir­rors the adapt­ab­il­ity observed in nat­ur­al evol­u­tion, where organ­isms refine their beha­viours over time to bet­ter res­on­ate with their environment.

Draw­ing inspir­a­tion from the archi­tec­ture of the brain, neur­al net­works in AI are con­struc­ted with nodes organ­ised in lay­ers that respond to inputs and then gen­er­ate out­puts. In the realm of human neur­al syn­chrony research, invest­ig­at­ing the role of oscil­la­tions has proven to be a pivotal area of interest. High-fre­quency oscil­lat­ory neur­al activ­ity stands out as a cru­cial ele­ment, demon­strat­ing its abil­ity to facil­it­ate com­mu­nic­a­tion between dis­tant brain areas. A par­tic­u­larly intriguing phe­nomen­on in this con­text is the theta-gamma neur­al code, show­cas­ing how our brains employ a dis­tinct­ive meth­od of ‘pack­aging’ and ‘trans­mit­ting’ inform­a­tion, remin­is­cent of a postal ser­vice metic­u­lously wrap­ping pack­ages for effi­cient deliv­ery. This neur­al ‘pack­aging’ sys­tem orches­trates spe­cif­ic rhythms, akin to a coordin­ated dance, ensur­ing the stream­lined trans­mis­sion of inform­a­tion, and it is encap­su­lated in what is known as the theta-gamma rhythm.

This per­spect­ive aligns with the concept of “neur­omorph­ic com­put­ing,” where AI archi­tec­ture is based on neur­al cir­cuitry. The key advant­age of neur­omorph­ic com­put­ing lies in its com­pu­ta­tion­al effi­ciency, address­ing the sig­ni­fic­ant energy con­sump­tion chal­lenges faced by tra­di­tion­al AI mod­els. The train­ing of large AI mod­els, such as those used in nat­ur­al lan­guage pro­cessing or image recog­ni­tion, can con­sume an exor­bit­ant amount of energy. For instance, train­ing a single AI mod­el can emit as much car­bon diox­ide as five cars over their entire lifespan⁸. Moreover, research­ers at the Uni­ver­sity of Mas­sachu­setts, Amh­erst, found that the car­bon foot­print of train­ing deep learn­ing mod­els has been doub­ling approx­im­ately every 3.5 months, far out­pa­cing improve­ments in com­pu­ta­tion­al effi­ciency⁹.

Neur­omorph­ic com­put­ing offers a prom­ising altern­at­ive. By mim­ick­ing the archi­tec­ture of the human brain, neur­omorph­ic sys­tems aim to achieve high­er com­pu­ta­tion­al effi­ciency and lower energy con­sump­tion com­pared to con­ven­tion­al AI archi­tec­tures¹⁰. For example, IBM’s TrueNorth neur­omorph­ic chip has demon­strated sig­ni­fic­ant orders of mag­nitude in energy effi­ciency com­pared to tra­di­tion­al CPUs and GPUs¹¹. Addi­tion­ally, neur­omorph­ic com­put­ing archi­tec­tures are inher­ently suited for low-power, real-time pro­cessing tasks, mak­ing them ideal for applic­a­tions like edge com­put­ing and autonom­ous sys­tems, fur­ther con­trib­ut­ing to energy sav­ings and envir­on­ment­al sustainability.

In the realm of train­ing and skill devel­op­ment, syn­chron­ised AI has the poten­tial to per­son­al­ise learn­ing exper­i­ences based on an employ­ee’s unique learn­ing curve, facil­it­at­ing faster and more effect­ive skill acquis­i­tion. From a cus­tom­er engage­ment stand­point, syn­chron­ised AI inter­faces might more pre­cisely under­stand and, in some cases, anti­cip­ate user expect­a­tions based on advanced beha­vi­our­al patterns.

For oper­a­tion­al effi­ciency, espe­cially in sec­tors like man­u­fac­tur­ing or logist­ics, AI sys­tems work­ing in coordin­a­tion with each oth­er can optim­ise pro­cesses, reduce waste, and strengthen the sup­ply chain. This would lead to increased prof­it­ab­il­ity, with an ever-met great­er abil­ity for sus­tain­ab­il­ity con­sid­er­a­tions integ­rated. In risk man­age­ment, syn­chron­ised AI sys­tems ana­lys­ing vast data­sets col­lab­or­at­ively might bet­ter pre­dict poten­tial risks or mar­ket down­turns, equip­ping busi­nesses and oth­er organ­isa­tions to pre­pare or pivot before a crisis emerges to lim­it all related social and soci­et­al impact. Like­wise, syn­chron­ised AI sys­tems could provide insights for more effi­cient urb­an plan­ning and envir­on­ment­al pro­tec­tion strategies. This could lead to bet­ter traffic man­age­ment, energy con­ser­va­tion, and pol­lu­tion con­trol, enhan­cing the qual­ity of life in urb­an areas.

In vari­ous domains bey­ond busi­ness, deploy­ment of AI with a proso­cial ori­ent­a­tion holds immense poten­tial for the well-being of human­ity and the plan­et. Par­tic­u­larly in health­care, syn­chron­isa­tion between the human brain and AI sys­tems could ush­er in a revolu­tion­ary era for patient care and med­ic­al research. Recent stud­ies high­light the pos­it­ive impact of clini­cians syn­chron­ising their move­ments with patients, thereby increas­ing trust, and redu­cing pain. Extend­ing this concept to AI chat­bots or AI-enabled robot­ic care­givers that are syn­chron­ised with those under their ‘care’ holds the prom­ise of enhan­cing patient exper­i­ence and improv­ing out­comes, as evid­enced by recent research indic­at­ing that LLMs out­per­formed phys­i­cians in dia­gnos­ing ill­nesses, and patients pre­ferred their interaction.

In the edu­ca­tion­al domain, the integ­ra­tion of AI sys­tems with a focus on syn­chrony is equally prom­ising. Research demon­strated that syn­chron­ized brain waves in high school classrooms were pre­dict­ive of high­er per­form­ance and hap­pi­ness among stu­dents¹². This study under­scores the sig­ni­fic­ance of neur­al syn­chrony in the learn­ing envir­on­ment. By lever­aging AI tutor­ing sys­tems cap­able of detect­ing and respond­ing to stu­dents’ cog­nit­ive states in real-time, edu­ca­tion tech­no­logy can poten­tially rep­lic­ate the pos­it­ive out­comes observed in syn­chron­ised classroom set­tings.  Incor­por­a­tion of AI sys­tems that res­on­ate with stu­dents’ brain states has the poten­tial to cre­ate a more con­du­cive and effect­ive learn­ing atmo­sphere, optim­iz­ing engage­ment and fos­ter­ing pos­it­ive learn­ing outcomes.

The excite­ment sur­round­ing the pro­spects of brain-to-machine and machine-to-machine syn­chrony brings with it a set of para­mount con­cerns that neces­sit­ate scru­tiny and that are all but tech­nic­al. Data pri­vacy emerges as a crit­ic­al appre­hen­sion, giv­en the intim­ate nature of neur­al inform­a­tion being pro­cessed by these sys­tems. The eth­ic­al dimen­sions of such syn­chron­isa­tion, par­tic­u­larly in the realm of AI decision-mak­ing, present com­plex chal­lenges that require care­ful con­sid­er­a­tion¹³¹⁴.

Expand­ing on these con­cerns, two over­arch­ing issues demand heightened atten­tion. Firstly, the pre­ser­va­tion of human autonomy stands as a found­a­tion­al prin­ciple. As we delve into the era of brain-machine syn­chrony, it becomes imper­at­ive to ensure that indi­vidu­als retain their abil­ity to make informed choices. Avoid­ing scen­ari­os where indi­vidu­als feel coerced or manip­u­lated by tech­no­logy is cru­cial in uphold­ing eth­ic­al standards.

Secondly, the ques­tion of equity in access to these tech­no­lo­gies emerges as a press­ing mat­ter. Cur­rently, such advanced tech­no­lo­gies are often costly and may not be access­ible to all seg­ments of soci­ety. This raises con­cerns about exacer­bat­ing exist­ing inequal­it­ies¹⁵. A scen­ario where only cer­tain priv­ileged groups can har­ness the bene­fits of brain-machine syn­chrony might deep­en soci­et­al divides. Moreover, the lack of aware­ness about these tech­no­lo­gies fur­ther com­pounds issues of equit­able access¹⁶.

The integ­ra­tion of AI with human cog­ni­tion marks the threshold of an unpre­ced­en­ted era, where machines not only rep­lic­ate human intel­li­gence but also mir­ror intric­ate beha­vi­our­al pat­terns and emo­tions. The poten­tial syn­chron­isa­tion of AI with human intent and emo­tion holds the prom­ise of rede­fin­ing the nature of human-machine col­lab­or­a­tion and, per­haps, even the essence of the human con­di­tion. The out­come of har­mon­ising humans and machines will sig­ni­fic­antly impact human­ity and the plan­et, con­tin­gent upon the guid­ing human aspir­a­tions in this pur­suit, and open oppor­tun­it­ies for an advanced human-centered AI exper­i­ence, in a “Fusion Mode”, as coined in the “Arti­fi­cial Integ­rity” concept. This raises a time­less ques­tion, rever­ber­at­ing through the course of human his­tory: what do we value, and why?

A cru­cial point to emphas­ise is that the implic­a­tions of syn­chron­ising humans and machines extend far bey­ond the realm of AI experts; it encom­passes every indi­vidu­al. This under­scores the neces­sity to raise aware­ness and engage the pub­lic at every stage of this trans­form­at­ive jour­ney. As the devel­op­ment of AI pro­gresses, it is essen­tial to ensure that the eth­ic­al, soci­et­al, and exist­en­tial dimen­sions are shaped by col­lect­ive val­ues and reflec­tions, avoid­ing uni­lat­er­al decisions by Big Tech that may not align with the broad­er interests of human­ity. What hap­pens next shapes our indi­vidu­al and col­lect­ive future. Get­ting it right is our shared responsibility.