https://data-lab.be/@DataLabBE
DemystifyingArtificial Intelligence
Prof. Tias Guns <[email protected]>http://homepages.vub.ac.be/~tiasguns/
19 Dec @UCIP China-EU AI Workshop
Tias Guns, VUB
Question
Is Artificial Intelligence:
A) a new type of human-created intelligence?
B) a research field in computer science that develops smart algorithms?
Tias Guns, VUB
Question
Is Artificial Intelligence:
A) a new type of human-created intelligence?
B) a research field in computer science that develops smart algorithms?
Dream
Reality
Tias Guns, VUB
Recent examples of A.I.
Tias Guns, VUB
Artificial Intelligence?
Predictions & Reasoning
To propose actions ( = decision support )
To take actions ( = autonomous system )
Tias Guns, VUB
From Big Data to A.I.
I. KNOW – Big Data= collect and analyse the now
II. PREDICT – Machine Learning= learn from the past to predict the future
III. REASON – A.I.= plan and act according to now and predicted future
Tias Guns, VUB
Can A.I. outsmart people?
YES a person is limited in space and time
an AI is limited in neither:parallel computing and the cloud
Tias Guns, VUB
Is A.I. Dangerous?
Tias Guns, VUB
Is A.I. Dangerous?
Tias Guns, VUB
Dangerous? Well, how is it made...
1) Define: a mapping from input to output
2) Collect: lots and LOTS of mapping examples
3) Science & Engineering: until the system produces satisfactory output from input
Tias Guns, VUB
Is A.I. Dangerous?
A.I. systems are savants“a person with a high level of knowledge or skill,
especially someone who is less able in other ways”(Definition of “savant” from the Cambridge Advanced Learner’s Dictionary & Thesaurus © Cambridge University Press)
No context, no moral, no knowledge of side-effects
→ it optimises one function, at any (unseen) cost
Tias Guns, VUB
Is A.I. Dangerous?
A.I. systems are savants“a person with a high level of knowledge or skill,
especially someone who is less able in other ways”(Definition of “savant” from the Cambridge Advanced Learner’s Dictionary & Thesaurus © Cambridge University Press)
No context, no moral, no knowledge of side-effects
→ it optimises one function, at any (unseen) cost
Tias Guns, VUB
A.I. in transport: congestion“The economics of Traffic Congestion”, R. Arnott, K. Small, 1994
Possible, road: optimize flow at traffic lights
Tias Guns, VUB
From Big Data to A.I.
I. KNOW – Big Data= collect and analyse the now
II. PREDICT – Machine Learning= learn from the past to predict the future
III. REASON – A.I.= plan and act according to now and predicted future
+ Potential dangers
Tias Guns, VUB
A.I. in transport: public transport optimisation
I. KNOW: train locations & how many people on it
II. PREDICT: delays & when full
III. REASON: priority rules, wait-at-stop policy, daily vehicle size, temporary vehicles, ...
+ DANGER: need cooperationfrom drivers and passengers,fairness of priority rules?
Tias Guns, VUB
A.I. in transport: freight truck planning
I. KNOW: where trucks/goods are and travel and delivery time
II. PREDICT: travel times, delivery times
III. REASON: which goods to which truck, routes to drive, loading of trucks, who drives which truck
+ DANGER: over-optimisation stresses out drivers, can break 'acceptable' way of working (human capital)
Tias Guns, VUB
A.I. in transport: urban freight management
I. KNOW: how do trucks move in the city
II. PREDICT: where will trucks enter and whereto
III. REASON: impact of road works or construction sites, manipulate flows with road restrictions, preferred routes
+ DANGER: other impact on stakeholders hard to measure
Tias Guns, VUB
Urban freight management
ViaPass OBU data: Belgian road tax for trucks > 3.5 ton GPS location updates every 30 seconds
collaboration with Brussels Mobility: understanding flows
https://data-lab.be/@DataLabBE
Conclusion
A.I. can make wishes come true.(after investments in infrastructure and data)
Let's wish for human-oriented,sustainable solutions!
What role can EU-China relations play here?How to overcome barriers in language and data regulation?