Presentation at Machine Intelligence 20 - Human-Like Computing Workshop.
http://alandix.com/academic/papers/mi20-human-like-2016/
There are clear connections between the long-term study of human–computer interaction (HCI) and the emerging area of human-like computing (HLC). A recent report on HLC to the HCI community identified four main topics: (i) improving interaction with people through human–like computation; (ii) developing new interaction paradigms for interacting with HLC agents; (iii) emulating human capabilities as a good model for general AI and robotics; (iv) learning more about human cognition and embodiment through HLC. This pa-per explores these topics focusing on the potential input of HCI knowledge and methods and using past and current HCI research as examples.
Axa Assurance Maroc - Insurer Innovation Award 2024
Seeking the Human in Human-Like Computing
1. Seeking the human in
human-like computing
Alan Dix
Talis and University of Birmingham
http://alandix.com/academic/papers/mi20-human-like-2016 /
Human-Like Computing Machine Intelligence (MI20-HLC)
2. clear intersections
Human-Like Computing
tentative EPSRC definition:
“offering the prospect of
computation which is akin to
that of humans, where
learning and making sense of
information about the world
around us can match our
human performance”
Human–Computer Interaction
studies the ways in which
people engage with
technology and the ways in
which technology can be
designed to work for and
alongside people
3. four topics
(iv) learning about
human cognition
(also social, ethical
consequences, etc.)
(iii) because it is a good
model to emulate
(new human-inspired algorithms)
(ii) new interaction
paradigms
(traditional focus has
human in control)
(i) for interacting
with people
(so need to understand
people!)
main goals for HLC (and HCI)
AI/ML – secondary benefits – HCI/CogSci
4. existing/past examples
(iv) learning about
human cognition
(iii) because it is a good
model to emulate
(ii) new interaction
paradigms
(i) for interacting
with people
1992 (!) black-box algorithms and
gender/ethnic discrimination
5. existing/past examples
(iv) learning about
human cognition
(iii) because it is a good
model to emulate
(ii) new interaction
paradigms
(i) for interacting
with people
cognitive and computational modelling
of regret and self
6. future challenges
(ii) new interaction
paradigms
(i) for interacting
with people
apply HCI knowledge,
theories and methods
for rich human
behaviour
build on existing
formal & informal
understanding of
human interaction with
active computational
partners
Editor's Notes
Alan is sorry not to be here in person to speak, but he sends his best wishes and is looking forward to catching up with the outcomes.
Alan’s own interest in this area stretches back more than 25 years, and this short presentation looks broadly at the connections between human-like computing and human-computer interaction illustrated by examples from Alan’s own work.
Human-like computing seeks to understand and develop ways in which intelligent algorithms and systems can operate in ways closer to a human; and human–computer interaction seeks to understand the ways humans work with technology and how systems can be designed to make this better.
There are clear intersections as recognised both by the EPRC human-like computing workshop earlier this year and the call for this workshop.
Alan reported about the EPSRC human-like computing workshop to the HCI community and identified four main topics or goals. Two of these, to interact more appropriately with people and develop new interaction paradigms for HLC, are the core aims of HLC and HCI’s contribution to HLC. The other two, human inspired algorithms and improving our understanding of human cognitive and social processes are additional secondary benefits to the AI/ML and HCI/CogSci communities respectively.
Existing work demonstrates some of the potential synergies. Very early work in applications of AI and pattern recognition in HCI (1992!) identified potential legal and ethical issues of black-box algorithms including gender and ethnic bias. However, this then led to investigating potential (partial) solutions including variants of ID3 and novel database querying techniques.
More recent work informed by issues of emotion in user interfaces, looked at cognitive models of regret and consciousness of self. This has implications for algorithm design, and in the case of regret has gone full circle, with improvements to machine learning and then lessons from this for qualitative understanding of emptions.
Focusing on the two core goal there are key future challenges. We need to apply the knowledge, theory and methods developed by 35 years of HCI research to help provide further the HLC agenda. However, perhaps most challenging will be to move beyond more instrumental modes of interaction design towards interaction where the computer has a more active role; this can build on existing work in areas such as virtual agents and human–robot interaction as well as more theoretical research including Alan’s work on formally modelling this kind of system.