- machine to (how machine can understand) - introduce machine human welfare - human (enhance human) sociological issue - pc dewey (simple is the best, but not good enough); k-12: skillfuture, smu academy (adult training) rsq: 1. why are there still so many jobs 1. how will we design tech and train humans to build skills and habits for human success in robots heavy future 2. adapt our social institution 2. how ai works, science of ai 3. is this time different (from previous revolutions)? impacts on society 4. what can we do? response from science, government, individual expert system (look up, knowledge based) vs learning system (imitate brain) sgd: 39.5million scared of ai: you are scared of other person able to use ai, rather than ai - converging force, rather than diverging force - meaning of the job david autor and daron acemoglou (automated tasks, non automated taks); complementarity and insatiability - frontier (ai specialist), wealth (gift wrapper, horse exerciser, sommelier, golf cart mechanic) , last mile job (vending machine attendant, -automation is not complete, teleprompter (half way to be replaced)) productivity vs human welfare shrinking middle class ``` Avg wellbeing index: ( 0.5 * "Average wellbeing from disposable income (1)" + 0.5 * "Average wellbeing from public spending (1)") * "Average wellbeing from inequality (1)" * "Average wellbeing from global warming (1)" * "Average wellbeing from progress (1)" ``` Jiangbo: what's the objective for human machine collaboration and society Bryan: how ai can improve middle class america Michael: should we prevent, how to prevent, - ntuc (employer, employee, government) Raha: 1. how to redesign job 2. enhance experience of autonomy and decision making for worker main trends, human footprint, consumption, wellbeing, turn around indicators, 10yr wave in labour market, 4 yr cycle inventory psychological adaption even for high education ppl are needed: how to translate 1d dna to (transcribe - protein); resist, resent, do not trust; deep mind alpha-fold beat big margin machine do not upgrade themselves; ai does, singularity (never teach ai how to program, access to internet) Q (Qingyi): ai via automation VS ai directly impact us: we can still benefit from automation (where to draw line) - how could the automation accept and tranition Q (Zhenzhen): how to make ai more accessible to lower wage (factory or construction worker) - how and what to train people Q. how humand and ai work together? - e.g. Changyi airport: airplane push back (4 people -> 1 people + machine, instead of dropping to zero), let the man do the main work and machines assist Q (Sherry Hou): how much should we learn on ai? - broader education; creativity in general is value, but the definition of creativity (e.g. type) may change - what advantage humans are better than ai Q (Nick ): Acemogouls's work political, economical share prosperity that (redistributive tax code) - is it worth looking back to history? - one of the professors are political science by training Q (Devin): how to tackle privacy issues? - we have a board: ethical, legal, environmental board (Sandy joining the team) - share insight without sharing data itself Q. (senior professor): median stabilized, mean continued growth: what role can ai play to make the tech impact more benign? - eduction (specific to people to college degree?) - Singapore has floor level (ulupandam study center (committee org. low income in Singapore; $100/month) vs Q (Bhuvan): how would we expect the future of job and prepare? increasing gap (e.g. among all the human call centers, only 2% get inspected - we can use AI to inspect all of them), welfare and - ai to screen quality of work, another generation of ai would get rid of last mile - salesman visiting 20-30 outlets -> increased production, recommendation improved sales Q (Awad): industrialization created subsequent job (does it replace?) make production and agriculture semi human replacement - last mile job is changing e.g. chatbot appeared - prof. Levi studied banking industry, and needed to delve into task level to analyze automation/replacement ## positive impacts - science, health, productivity, eudcation, - job displacement (malicious use of ai - e.g. scam), job placement, disinformation and fate of democracy (Harali - conversation is the core of democracy), super-intelligence out of control ## response from sci, gov, individual - can we control sth smarter than us? Ruseel Stuart machine are intelligent if their, danger comes from machine having single minded goal (machines should have explicitly uncertain about human preference) - compute pi, max tegmark - machanistc interpretability - neurons in geometric space (introduce cost - distance btw two neurons - penalized further away - encourage small modules) capability vs alignment ai needs its own ligo moment (large scale collaboration) what are we still better at dexterity, social intelligence, creativity, long term planning creativity, communication, critical thiking training (skills, jobs, careers) vs education (enlightenment) most important skilled and flexible Singapore advantage: education and competent, unapologetic social engineer (racial quota - neighborhood) - certifacte of, aggressive public policy intervention 90s: mit teaching singapore 00s: smart (research, ) thank you mit, we don't need you 10s: can you bring integrated enterprises + innovation center Singapore need ppl (need AI automation) one generation of people lose their job, textile industry (retrain me to do other thing) --- - - gpt is able to do long term planning