Future of AI


  • Authenticity of the report
  • What is AI?
  • AI research trend
  • AI in domain
  • Success of AI
  • Problems created by AI
  • Recommendations
  • Questions unanswered
  • AI in Team Compiler



  • First in series "One Hundred Year Study on Artificial Intelligence"
  • Report both reflects on progress done in last fifteen years and anticipates progress in next fifteen years


AI can be defined as

Artificial Intelligence (AI) is a science and a set of computational technologies that are inspired by—but typically operate quite differently from—the ways people use their nervous systems and bodies to sense, learn, reason, and take action

AI in 19s

  • Progress was patchy and unpredictable
  • Lack of Data
  • Lack of Computing power

21st century AI

  • AI enables constellation of mainstream technologies having impact on everyday life

AI Research Trend

Shift in field from building intelligent systems to human-aware and trustworthy intelligent systems

AI backbone

  • Innovations in Hardware
    • Sensing
    • Perception
    • Object sensing
    • GPUs (How can we miss this?)
    • DL focused hardware
      • TPUs 
      • Nervana

Large scale Machine Learning

  • well understood (Un) Supervised learning.
  • Current Focus is scaling existing algorithms to work with large data sets
  • Open source libraries
    • Mlpack, mlpy, pyml, caffe and many more

Deep learning

  • Training Convolutional neural network
  • In Computer Vision
    • Object recognition
    • Video labeling
    • Activity recognition
  • In perception
    • Audio, Speech and NLP

Reinforcement Learning

  • Shifts focus to Decision making
  • Experience-driven
  • No success in past due to issues of representation and scaling
  • Deep learning is NOS to RL
    • Success of AlphaGo - program by Google DeepMind
      • Initially trained with human expert database
      • then applied RL by playing with itself (Computer programs are not biased)


  • Static environment is solved
  • Current efforts in training robot to interact with world in generalizable and predictable ways.

Computer Vision

  • Most prominent form of machine perception
  • Emerged after DL
  • Use of Support Vector Machine
  • Large data set, GPUs, refinement of neural network algorithm

Natural Language Processing

  • 20% of current mobile queries(Android) are done by voice
  • Real time translation
  • Current focus on interactive dialogue with humans 
  • WaveNet - Mimics human voice
    • https://deepmind.com/blog/wavenet-generative-model-raw-audio/

Collaborative systems

  • Help develop autonomous systems that works collaboratively with other system and with human.
  • works on complementary strengths of human and machines


  • Array of devices interconnected to collect and share their sensory information
  • includes appliances, vehicles, buildings, cameras and lot more
  • Current
    • Drone
    • Nvidia Jetson

Neuromorphic Computing

  • VLSI systems containing electronic circuits to mimic neuro-biological architecture present in nervous system
  • No big win yet.
  • Qualcomm
    • https://www.technologyreview.com/s/526506/neuromorphic-chips/

Quantum Computing

  • learning rate will come down to seconds
  • http://www.dwavesys.com/


  • DL influencing with labeled data
  • Reinforcement learning obviates the requirement of labeled data
  • Machine perception, computer vision, force and tactile perception.


1. Transportation

  • Smarter cars
    • Drive CX
  • Self-driving cars
    • Drive PX
  • Transportation planning
  • Less traffic
  • Utilization of travelling time  

2. Home/Service Robots

  • Nothing to say

3. Healthcare

  • Clinical decision support
  • patient monitoring and coaching
  • mining social media to infer possible health risks
  • Robotics for surgery
  • Challenge - Poor human-computer interaction
    • Perception will help bridge gap
  • Electronic Health Record
  • Healthcare analytics
    • AI program suggested surgery earlier declared as impossible by expert surgeon
  • Cancer and all other diseases
  • Smart devices

4. Education

  • Teaching robots
    • http://ozobot.com/
  • Intelligent Tutoring System and online learning
  • Simulating problems in aircrafts
  • MOOCs 



5. Low-resources communites

  • Fear of AI contributing to joblessness
  • Predictive models helping government agencies to effectively use budget to address public problems
    • Illinois Department of Human Services using predictive models to identify pregnant women at risk
    • The City of Cincinnati identifies and deploys inspectors to properties at risk of code violations.

6. Public safety and security

  • Surveillance 
  • White collar crimes - Credit Card fraud
  • Cybersecurity
  • The New York Police Department’s CompStat
    • Step towards predictive policing
    • https://compstat.nypdonline.org/

7. Employment and workplace

  • AI will replace tasks rather than jobs
  • Will create new types of jobs
    • Now Machines assists humans. Then, Humans will  assist machines.
  • Cognitive jobs will be affected just like automation, robotics affected manufacturing
  • Labor becomes less important factor in production

8. Entertainment

  • Already used in Movies, Games
  • https://www.wordseye.com/
  • Improve/edit videos and images captured by average camera
    • Pix by Microsoft
    • Prisma (I also said edit)

Success of AI

The ease with which people use and adapt to AI applications will determine success

  • Adaptation of Self-driving cars
  • Interactions with everyday things like Oven, TV, Fridge, and much more

Problems created by AI

1. Mistakes will diminish trust from AI systems

2. Effect on peoples capabilities

  • Calculator in classroom has reduced childrens capability for basic arithmatics

3. More reliance on personal assistance and virtual agents deteriorating thinking for self

4. Unavailability of system for all

Content in Chinese










5. Possibility of Biases

  • System design can incorporate biases in data sources or the way system learns.
    • Similar to how kids learns.


1. Technical expertise in AI at all levels of government

  • Official should to assess safety and other metrics of AI system
    • Human behavior and societal values will have impact.

2. Remove perceived and actual impediments to research impact of AI system 

3. Increase public and private funding for interdisciplinary studies

Questions unaddressed 

1. Who will be responsible when self-driving car crashes?

2. Preventing unlawful discrimination of AI application

  • Example, Microsofts twitter bot

3. What protection should be afforded to people whose skills are rendered obsolete due to AI?

AI in Compilers

1. EVO

  • Phase Ordering
  • Dead shaders elimination

2. Updating heuristics on the fly

  • Heuristics values (approximated) can be tuned as per app on the fly
  • Perf and Compile Time benefit
    • CT Example - Heuristic for minimizing register target

1. Automated test creation

2. Minimize test cases

  • 80-20 principal
  • Analyze pattern in real apps
    • Tegra - Games
    • Compute - Cuda apps
  • Keep tests with similar pattern

Future of AI

By Bhushan Sonawane