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.