AI implicit costs Flashcards
(10 cards)
AI implicit costs
- technical issue
“Bigger is better” → increasing parameters
Too many parameters
- too big
- less quality of the sources
New approach
- less sources but good quality
- less decimal
AI implicit costs
- Impact on the environment
→ Power demands
- 2026 consume 100 times of 2023
→ Carbon footprint
- Data centers power consuming infrastructures
- AI more consuming then traditional search
→ Heuristic processes and not analytic mental process
(we could think before of the impact without trial)
→ Biases
AI implicit costs
- Heuristic process bias
1) WYSIATI
2) The intentional stance toward the AI
3) Somatic markers
4) Hyperbolic discounts
5) Herd and authority bias
6) Isolated action bias
AI implicit costs
- Heuristic process bias
1) WYSIATI
- What you see is all there is
→ focus exclusively of
- what is visible here and no
- what we always know
-what seems true
→ overlooked
- less-knomw
- hard-to find
- hard-to understand
→ hide ethically relevant costs
- workers
- land
- energy
- etc.
→ cultural factors
- virtual element of AI
- not true or relevant
AI implicit costs
- Heuristic process bias
2) The intentonal stance toward the AI
→ AI as rational agent
- with belief and desire (Intention)
pro → it helps us explain, interact
cons → social and environment costs
→ helps spread disinformation about AI
AI implicit costs
- Heuristic process bias
3) Somatic markers (Sms)
→ physiological states that operate a selection between positive and negative stimuli
- affect decision-making
→ we don’t have direct experience of the consequences of climate-change
- Sms do not activate
AI implicit costs
- Heuristic process bias
4) Hyperbolic discounts
→ Future advantage (or costs) look smaller than real
AI implicit costs
- Heuristic process bias
5) Hard and authority bias
social media → echo chamber
- conspiracy theories
AI implicit costs
- Policies for a “Green” AI
Intervention
Intervention on researchers
→ Developing alghorithms, methodologies, system arrchitecture more:
- efficient
- energy-saving
- environmental-friendly
AI implicit costs
- Policies for a “Green” AI
Good practices
→ Short-term benefits to develop green AI
- tax-cuts
- financial rewards
(hyperbolic discounts)
(isolated action bias)
→ reward researchers that embrace green AI
- activate somatic marker
→ implementing the measures on a massive scale
- active herd and the authority bias