We believe that machine intelligence and robots will eventually be capable of performing all human jobs.
Machine learning and robotic dexterity have in many cases already surpassed human abilities. The labor economics for productivity have blurred; it is increasingly difficult to differentiate between human and machine productivity.
Based on our blog “Imitating Machines”, we will speak frankly about the realities of AI and machine sapience, including overfit models, creativity, understanding, intent and freewill.
AI and robotics can compete with people for all jobs. Remember, AI doesn't feel pity, remorse, or fear. And AI will not stop improving. Ever. So what should society do?
1. Why Your Job Is
Not Safe
From Robots and AI
SXSW Interactive 2016 PanelPicker Proposal
http://bit.ly/AIandJobs
Paul Teich
Sam Teich
2. We are not going to concentrate on:
•Sentience and self-awareness
•Singularity, abundance, or other
features of Kurzweilian futures
•Evil killer robot/AI apocalypse a la
Gates, Musk, Hawking, etc.
3. Our approach is conservative
How did history lead us to where we are today?
What are the technically possibilities for your product
by 2020?
4. •Makes it easier for one person to perform a job faster
•Which means fewer people are needed to perform the
same amount of work
We believe it is possible
that robotics and “thinking”
machines will eventually
take all human jobs –
But certainly not by 2020
Automating a Human Task
5. Agricultural Revolution
Created a surplus of food
•Which created a surplus
of time
Enabled people to invest in
life-improving crafts and
complex social hierarchies
6. Industrial Revolution
Automated repetitive physical work
•From manufacturing to moving
things and people
Reduced the number of workers
needed to produce and deliver physical
goods and services
Displaced mostly
•Unskilled muscle-based labor
•Repetitive, rote physical labor with
no decision-making component
7. IT Revolution
Automated repetitive mental work
•Spreadsheets to decision support systems
Reduced the number of workers needed to produce and deliver
any kind of repeatable goods and services
Displaced mostly
repetitive, rote tasks
• Transcription-based labor
• Arithmetic and process-
based labor
8. Synthesis Revolution
Combine sensory input and robotic dexterity with feedback loops,
deep learning and artificial intelligence
Potential to displace broader worker roles in producing and
delivering adaptable goods and services
Automate non-repetitive sensory
feedback dependent physical work
• Visual
• Auditory
• Tactile
• Olfactory and Gustatory are also on the way…
9. MOOV - Machine Creativity
Create complex models of reality that can
generate occasionally false—but useful—
understanding
Predict multiple outcomes describing different
futures
Have opinions about which outcomes to pursue
Demonstrate volition—“want” to change the future
Resulting in new and unique goods and services...
10. We’ve talked about the big picture
Which specific tasks are or shortly will be
automatable?
How will this affect employment?
11. Partial Automation
It isn’t necessary to automate every task
composing a job for some people to
become unemployed due to automation.
The incremental increases in productivity
automation brings may lead to excess
labor supply.
12. Tasks and Automation
● Driving
● Complex Assembly
● Janitorial Services
● Plumbing
● Forming/Testing Hypotheses
● Medical Diagnosis
● Legal Writing
● Persuasion
● Management
● Picking or Sorting
● Repetitive Assembly
● Record-keeping
● Calculation
● Repetitive Customer Service
Autor,LevyandMurnane,2003
Many tasks which economists assumed to be non-routine are increasingly found to be
routine and therefore more susceptible to machine learning tasks than previously thought.
Analytic and Interactive
RoutineNon-Routine
Manual
13. Airbus Open Robotic Cell and Robot Interface
Wide range of jobs, physical
environments and product layouts
Remove language barriers between…
•Humans
•Machines
•Information systems
Robots working alongside humans in major component assembly &
final assembly line to execute human tasks without human limitations
14. “Lump of Labor” Fallacy
Definition:
Demand for work is fixed, therefore
Increases in worker productivity reduce the job pool
(variant of zero-sum game)
We believe:
Demand for work is not fixed
But…
During periods when worker productivity increases
Faster than demand for work
The net effect is job losses
15. Virtuous and Vicious Cycles
Success Investment
InnovationProduction
$Rev $Risk
Automation
Job Gain/Loss
for Production
Job Gain/Loss
for Innovation
Net
Labor
Flow
$ Gain/Loss for
Production
$ Gain/Loss for
InnovationNet
Capital
Flow
Increasing
Automation may
lead to imbalances
in Labor Supply
and Demand, and
divorce Capital
from Labor.
16. What Can We Do?
Say ‘no’ to progress? Really?
Create “education offsets”?
Stall progress through lobbying
and politics?
Legislate a mix of human and non-
human labor?
This is a tough challenge...