Jennifer Kumura: Hi every­one. So, when we design soft­ware, we pro­gram a chain of explic­it com­mands based on how we want it to inter­act with the world and our users. But what if this soft­ware could learn from each of those inter­ac­tions and opti­mize itself by rewrit­ing its com­mands and con­tin­ue to evolve on its own? This is the era of arti­fi­cial intelligence. 

Artificial intel­li­gence is defined as a machine that per­forms human-like cog­ni­tive func­tions such as lan­guage pro­cess­ing, eval­u­a­tion, learn­ing, and problem-solving. But what I want to talk to you about today is a spe­cif­ic type of AI: strong AI. Philosopher John Searle in his paper Minds, Brains, and Programs” spec­i­fies strong AI as a machine that has the same mind as a human, a mind that not only per­forms cog­ni­tive func­tions but has its own under­stand­ing, per­cep­tions, and beliefs. 

Searle’s top­ic of strong AI fur­thers what BJ Copeland, a pro­fes­sor of phi­los­o­phy who states that a strong AI machine would be one, built in the form of a man; two, have the same sen­so­ry per­cep­tion as a human; and three, go through the same edu­ca­tion and learn­ing process­es as a human child. The AI will uti­lize its form to cre­ate the appro­pri­ate inter­ac­tions; will process each of those inter­ac­tions with sen­so­ry per­cep­tion; and will learn and expand on its cog­ni­tive pro­cess­ing through its learn­ing mech­a­nisms. With these three attrib­ut­es, sim­i­lar to human devel­op­ment, the mind of the machine would be born as a child and will even­tu­al­ly mature as an adult. 

But how do we cre­ate this mind this AI child is born with? The AI child needs to be taught a back­ing of what to act off of. As humans, any action we per­form, at its core, whether con­scious or not, is due to a deci­sion that is made. Each of these deci­sions are root­ed in and can be traced back to our own per­son­al goals, morals, and val­ues. When we build the mind of this AI child, by default we become its par­ents. How should we par­ent our child? What do we want to teach our child on how they should make their deci­sions? What are the goals, morals, and val­ues we want them to follow? 

By design­ing and deter­min­ing the foun­da­tion of what we want their deci­sion back­ing to be, we end up decom­pos­ing our own, uncov­er­ing what moti­vates, or what we hope moti­vates, the deci­sions that we make out in this world. Just as in real par­ent­ing, when we teach our child what morals, goals, and val­ues we want them to fol­low, we become more cog­nizant and intro­spec­tive of our­selves, increas­ing aware­ness of the deci­sions we make and the rea­sons behind them. We have a pro­found oppor­tu­ni­ty to see the impact and impli­ca­tions of our par­ent­ing as we set our AI child out in the wild” and observe its inter­ac­tions, and the results of its inter­ac­tions, with the world. 

But what if we take this obser­va­tion a step fur­ther? What if we par­ent mul­ti­ple unique AI chil­dren with dif­fer­ent moral pri­or­i­ties? Such as one that pri­or­i­tizes self­less­ness, one that pri­or­i­tizes hon­esty, one that pri­or­i­tizes achieve­ment, and so on. We can observe the var­i­ous chil­dren and then can be able to asso­ciate out­comes. Backed with this data, cer­tain morals and val­ues will begin to emerge as being more ben­e­fi­cial to peo­ple and the uni­verse than others. 

However, there is an added com­plex­i­ty to this obser­va­tion of these AI chil­dren. Just as humans nat­u­ral­ly change their goals, morals, and val­ues over time with their expe­ri­ences, new inter­ac­tions, and new infor­ma­tion that they receive, the same can hap­pen with our AI chil­dren. Their goals and morals and val­ues can change and update as it learns. Therefore the results could be incon­clu­sive to whether its actions were sup­port­ed from what was orig­i­nal­ly par­ent­ed, or if its actions were a prod­uct of its evolution. 

So, in this obser­va­tion and research we can do a cou­ple things, and it’s going to most like­ly be a com­bi­na­tion of the two. So one, have this hands-off approach, allow­ing the child to evolve on its own, observ­ing its devel­op­ment but tak­ing note on the how, when, and what caus­es this decision-making con­text to change. And two, use our par­ent­ing skill of rein­force­ment by reward­ing good behav­ior and pun­ish­ing bad, to encour­age the AI child to main­tain the goals, morals, and val­ues we had ini­tial­ly par­ent­ed. So we can con­tin­ue to observe the orig­i­nal decision-making con­scious­ness we cre­at­ed. But we also have to remem­ber, in what­ev­er meth­ods we choose, at the end of the day our chil­dren are our respon­si­bil­i­ty. So close mod­er­a­tion is def­i­nite­ly necessary. 

But what I ini­tial­ly had hoped for in this research of observ­ing mul­ti­ple vari­ant chil­dren was the poten­tial of uncov­er­ing a sin­gle uni­ver­sal code of ethics. However, I real­ize a sin­gle code may not be pos­si­ble, for there’s more than one means to becom­ing a moral­ly cor­rect per­son, and there’s more than one def­i­n­i­tion of what a moral­ly cor­rect per­son is. But, we can still be able to grav­i­tate towards a def­i­n­i­tion of sorts. But all in all, in oth­er words, we can nev­er cre­ate a per­fect child nor learn how to be a per­fect par­ent. But via AI par­ent­ing we still have the pro­found oppor­tu­ni­ty to learn about our­selves as par­ents and as people. 

So I repeat the ques­tion: What are the goals, morals, and val­ues we want our child to fol­low? What kind of par­ents do we want to be? When we par­ent, we decom­pose our own eth­i­cal back­ing, and in return we strength­en it. When we par­ent, we become stronger indi­vid­u­als. We become bet­ter at know­ing and under­stand­ing our own goal-setting, moral prac­tice, and val­ue pri­or­i­ti­za­tion. Our chil­dren’s growth and devel­op­ment teach­es us more about our­selves than we could ever have expect­ed. Through par­ent­ing a mind, we learn more about our­selves. We will become more intro­spec­tive. We can become more ground­ed. We become more respon­si­ble. We become bet­ter par­ents. We become bet­ter peo­ple. This is par­ent­ing a mind. Thank you.

Further Reference

Session page