Advanced Artificial Intelligence
Download PDF
Advanced Artificial Intelligence
Use the future to build the present
Advanced Artificial Intelligence
Comment
Stakeholder Type
5.5SyntheticBiology5.4Science ofthe Originsof Life5.3FutureEconomics5.2Future ofEducation5.1ComplexSystemsScience4.4Democracy-affirming Technologies4.1Science-basedDiplomacy4.2Advances inScience Diplomacy4.3Digital Technologiesand Conflict3.7InfectiousDiseases3.6Solar RadiationModification3.5OceanStewardship3.4SpaceResources3.3Future FoodSystems3.2WorldSimulation3.1Decarbonisation2.6FutureTherapeutics2.5Organoids2.4ConsciousnessAugmentation2.3RadicalHealthExtension2.2HumanApplicationsof GeneticEngineering2.1CognitiveEnhancement1.6CollectiveIntelligence1.5AugmentedReality1.4BiologicalComputing1.3Brain-inspiredComputing1.2QuantumTechnologies1.1AdvancedAIHIGHEST ANTICIPATIONPOTENTIAL
5.5SyntheticBiology5.4Science ofthe Originsof Life5.3FutureEconomics5.2Future ofEducation5.1ComplexSystemsScience4.4Democracy-affirming Technologies4.1Science-basedDiplomacy4.2Advances inScience Diplomacy4.3Digital Technologiesand Conflict3.7InfectiousDiseases3.6Solar RadiationModification3.5OceanStewardship3.4SpaceResources3.3Future FoodSystems3.2WorldSimulation3.1Decarbonisation2.6FutureTherapeutics2.5Organoids2.4ConsciousnessAugmentation2.3RadicalHealthExtension2.2HumanApplicationsof GeneticEngineering2.1CognitiveEnhancement1.6CollectiveIntelligence1.5AugmentedReality1.4BiologicalComputing1.3Brain-inspiredComputing1.2QuantumTechnologies1.1AdvancedAIHIGHEST ANTICIPATIONPOTENTIAL

Emerging Topic:

1.1Advanced Artificial Intelligence

    Associated Sub-Fields

    Artificial Intelligence (AI) aims to build machines that are able to behave in ways we associate with human activity: perceiving and analysing our environment, taking decisions, communicating and learning. There are various approaches to achieving this. The most well-known, and arguably most advanced, is machine learning (ML), which itself has various broad approaches.

    Show advancements in the past year

    To mention just two approaches, in supervised learning algorithms make associations between a given input and the desired output by learning on training sets comprising many correct input/output pairs. In reinforcement learning, the ML algorithm repeatedly chooses from a given set of actions in order to maximise a reward function which should lead it to the desired result. A typical example is learning to play a game such as Go, chess or video games, where the reward function is increasing the score or winning the game. Reinforcement learning is considered to be a promising strategy to address complex real-world problems.

    Machine learning algorithms have passed a number of impressive milestones in recent years. They identified objects by vision better than humans in 2015.1 The following year, they beat a Go champion and started playing complex video games.2 Autonomous cars have driven tens of millions of kilometres with very few accidents.3 Deep learning algorithms have become extraordinarily adept at mimicking traditionally human activities such as language processing, artistic creation and even scientific research.4 This rapid and impressive progress is primarily due to the increasing amount of available data and computing power. However, many applications require even more sophisticated skills, such as the ability to make sensible decisions in highly uncertain environments; transparency and traceability; the ability to combine data from highly heterogenous sources, and long-term memory and the inclusion of context.

    Selection of GESDA best reads and key reports

    There are several large-scale efforts to map the state of the art of artificial intelligence and to predict its evolution. Stanford’s “One Hundred Year Study on Artificial Intelligence” produces a summary of the major technological trends and applications by domains as well as legal, ethical and policy issues every five years.5 The “20-Year Community Roadmap for Artificial Intelligence Research in the US” from the Association for the Advancement of AI (AAAI) proposes detailed research roadmaps and recommendations about research infrastructures and education.6 The yearly State of AI Report summarises the main developments of AI of the past year in the field of research, industry and politics as well as education and experts.7 Other roadmaps focus on the opportunities and challenges of integrating AI in government, society and industry from European8 and Chinese9 perspectives.

    Emerging Topic:

    Anticipation Potential

    Advanced Artificial Intelligence

    Sub-Fields:

    Deeper Machine Learning
    Human-centred AI
    Next-level AI
    Interdisciplinary AI
    Rapid progress in the development and adoption of AI has already spurred many efforts to chart the trajectory of this transformative technology. While advances in deeper machine learning could have widespread consequences, it has already received considerable focus and has a relatively short path to maturity, so the need for anticipation here is lower. Next-level AI --- and the path towards human-level intelligence --- on the other hand, has received less attention so far and is further from maturity, suggesting more work is needed to understand its potential implications.

    GESDA Best Reads and Key Resources

    Article

    Why AI is Harder Than We Think // 28.04.2021, arXiv

    Published:

    7th Aug 2021
    Since its beginning in the 1950s, the field of artificial intelligence has cycled several times between periods of optimistic predictions and massive investment (“AI spring”) and periods of disappointment, loss of confi- dence, and reduced funding (“AI winter”). Even with today’s seemingly fast pace of AI breakthroughs, the development of long-promised technologies such as self-driving cars, housekeeping robots, and conversational companions has turned out to be much harder than many people expected. One reason for these repeating cycles is our limited understanding of the nature and complexity of intelligence itself. In this paper I describe four fallacies in common assumptions made by AI researchers, which can lead to overconfident predictions about the field. I conclude by discussing the open questions spurred by these fallacies, including the age-old challenge of imbuing machines with humanlike common sense.