“Can Bots be Better Learners than human?” is discussion paper initiated by Wassim Derguech and Mathieu d’Aquin in the context of re-coding black mirror workshop at ISWC 2017.
The objective of the paper, is not to find an answer to this question but rather to discuss ethical issues around the use of technology in the context of learning analytics.
In this context, the disuccion starts from the emerging trends in the use of chat bots for answer cutomers queries in online pltforms and the use of online pltaform in learning analytics.
1. Can Bots be Better Learners than Humans?
Wassim Derguech and Mathieu d’Aquin
Insight Centre for Data Analytics
National University of Ireland, Galway
Workshop@ISWC2017
2. Chat bots for communicating
with human users
2
[1] Gartner: http://www.gartner.com/newsroom/id/3551217
Img credit: https://yourstory.com/2016/11/68224f2e70-chat-bots-customer-support-channel-for-digital-age/
• By 2019, 20% of user interactions with smart phones will take place using a virtual assistant (a bot) [1]
• Chat bots use knowledge bases with static or dynamic patterns to answer human queries
• Knowledge bases can be enriched with new conversations/resources
3. “Be Right Back” in Black Mirror
3
EP01S02
• An artificial program is
used to simulate the
behavior of a deceased
person
• Input: online social
interaction
The Web a learning
platform to create the
artificial knowledge of bots
4. Learning Analytics
4
According to Wikipedia (and past LAK CFPs, and some papers from relevant people)
Learning analytics is the measurement, collection,
analysis and reporting of data about learners and their
contexts, for purposes of understanding and optimizing
learning and the environments in which it occurs.
5. Learning Analytics
5
According to Wikipedia (and past LAK CFPs, and some papers from relevant people)
Learning analytics is the measurement, collection,
analysis and reporting of data about learners and their
contexts, for purposes of understanding and optimizing
learning and the environments in which it occurs.
6. Learning Analytics
6
According to Wikipedia (and past LAK CFPs, and some papers from relevant people)
Learning analytics is the measurement, collection,
analysis and reporting of data about learners and their
contexts, for purposes of understanding and optimizing
learning and the environments in which it occurs.
not only humans
not only the classroom,
university, library or VLE
7. Learning Analytics: Basic Concepts
7
Learner
Platform
Analytics
VLE | Website | Library
Assessment | Enrollment
School/University
Prediction Drop out
BI
Planning
Recommendation
A university uses data on the
students and their activities
collected through the
institution's information systems
with the goal to predict their
success so they can improve
help them improve….
… and improve their teaching,
offering, environments as well
8. Learning Analytics: Everyday Learning
8
Learner
Platform
Analytics
VLE | Website | Library
Assessment | Enrollment
School/University
Prediction Drop out
BI
Planning
Recommendation
9. Learning Analytics: Everyday Learning
9
Learner
Platform
Analytics
VLE | Website | Library
Assessment | Enrollment
School/University
Prediction Drop out
BI
Planning
Recommendation
10. 10
Learner
Platform
Analytics
VLE | Website | Library
Assessment | Enrollment
School/University
Prediction Drop out
BI
Planning
Recommendation
Sentiment Analysis
Collective Intelligence Behaviour Analysis
Collaboration
Community Support
Learning Analytics
“Learn how to learn”
11. 11
Learner
Platform
Analytics
VLE | Website | Library
Assessment | Enrollment
School/University
Prediction Drop out
BI
Planning
Recommendation
Sentiment Analysis
Collective Intelligence Behaviour Analysis
Collaboration
Community Support
Learning Analytics
Assisted by Bots
14. Any Objection?
• Learning Analytics captures the cognitive process
from a surface point of view
• Learning happens as a side effect of interaction
between cognition system of the learner and the
social system in which they operate (co-evolution
model)
14
15. A Step too Far?
• Learning Analytics, as currently used by educational
institutions, looks at particular indicators to analyze and
predict the students’ performances
• Learning practices can be tailored for maximizing those
indicators “act” like good learners
• Bots can be programmed to maximize such indicators
15
Dear bot,
1. Visit all pages!
2. Spend 1 minute
per line.
3. Like all images.
…
16. Conclusion
• Two main trends: bots and learning analytics
• What if they merge together?
– Learning bots
– Learning assistants
• Objections:
– Learning analytics look at learning from a surface point of
view
– The learning happens as a side effect of learning interactions
• A step too far:
– Learning analytics looks at some indicators
– Bots can be used to maximize those indicators and fake the
learning process
16
Hinweis der Redaktion
“Can Bots be Better Learners than human?” is discussion paper initiated by Wassim Derguech and Mathieu d’Aquin in the context of this workshop
The objective of this paper, is not to find an answer to this question but rather to discuss ethical issues around the use of technology in the context of learning analytics.
In this context, the disuccion starts from the emerging trends in the use of chat bots for answer cutomers queries in online pltforms and the use of online pltaform in learning analytics.
Indeed, it has been clearly noticed in the last decade, a wide adoption of support services in using chat bots [next slide]
That are suitable for answering customer queries in a cost effective and efficient manner.
To give an indication of the wide adoption of bots in such context, Gartner predicts that by 2019, 20% of the user interactions with smart phones will take place using a virtual assistant or a bot.
Bots use either a static of dynamic knowledge base for funding the correct answer to a query or maintain a conversation with a human user.
This knowledge base can be continuously updated and enriched using new conversations or new textual and multimedia resources to help the bot “learn” how to answer a new question, propose a decision or compete with humans (e.g., bots in games)
Resources of knowledge can include the web in general.
Not far from this idea, in the “Be right back” episode of the second season of black mirror, an artificial program is used to simulate the behavior of a deceased person from his online social interactions.
On the other hand, there is the trend of using the web as a learning resource through online platforms and tools. [next]
According to Wikipedia ….
From this definition, two main concepts need to be highlighted:
The learner and his learning
And the learning environment
Learners are not necessarily human learners, this can also include bots and the learning environment can go beyond the classical entrainments such as classrooms, universities, etc. and include online social platforms such as facebook, twitter, youtube, etc.
Typically, learning analytics as it is used by universities consists of using data … [read from slide]
Furthermore, a an every day learning manner, users can also use other online resources such as social media platforms
In such context, even non academic learners (i.e., students) can be also considered as they are also grasping knowledge from these social platforms and other online resources.
Trends in every day learning as it is suggested in the AFEL project, propose to give users feedback on their performance in learner as well as teach them how to learn.
This can be achieved through collaborations with other learners, suggestion of learning material based on their progression ,etc.
I f we try to link both these trends (i.e., use of bots as personal assistants and learning analytics), one can image having bots that would simulate the human learning behavior to lean.
They can even increase the learners’ performance by suggesting learning material, summarizing long texts, evaluating contents, etc.
To a larger extent, bots themselves can “learn” from humans how to “learn” or enrich their knowledge bases.
In other words if if we consider the objective of the bot is to integrate knowledge from a certain domain in order to better answer an inquiry or any other task, we can imagine the bots summing up the human learning experience that can be captured in a learning analytics model. [next]
And this is not science fiction, this what is happening with technology advances and newly developed AI algorithms, etc.
However, when it comes to learning, Things might be more challenging for bots.
Bots are predictable, and they behave the way they are programmed, while humans are not.
Human can react differently to the same situation, take longer time to ingest data and appreciate information, etc.
This makes the learning “trajectories” of bots to be more programmable: search data, analyze it, identify millions of facts, etc in a blink of an eye.
However, they cannot understand human emotions that are critical in problem solving and performing tasks across domain.
Emotions have a considerable impact on learning performance. A bot that understands and simulate human emotions during learning trajectories remain challenging.
Learning Analytics captures the cognitive process from a surface point of view, indeed, analytics tool might report on what a learner has seen, for how long, how often, etc.
They cannot properly capture what the earner has actually learnt.
As suggested by many theories of models of learning (e.g., the co-evolution model): Learning happens as a side effect of interaction between cognition system of the learner and the social system in which they operate.
To some degree, this involves human emotions that we discussed earlier.
The main objective of learning analytics is to analyze and predict learners’ performances on the basis of some metrics that are identified from the learner interaction with an learning platform.
Example of metrics can include the pages visited, the flow of topics explored, the time spent on a certain page, the posting of questions and answers in a the courses forums, etc.
To a certain degree, one can adopt a learning practice that aims to maximize those indicators. In the same trend, one can program an assistant learning bot to “act” like a good learner to influence the learning indicators.
What was presented in this paper is a discussion regarding the merge of two main technology trends: bots and learning analytics.
Two potential scenarios can exist in this merger:
making bots learn by themselves to enrich their knowledge bases to find the right answers to queries, help making decisions or compete with humans in games.
Using bots as personal learning assistants to help identify the right learning materials, summarizing them, suggesting new ones, etc.
The paper discusses also the main objections to merge these technologies:
In fact, learning analytics look at learning from a surface point of view while the actual learning and acquisition of knowledge happens as a side effect of learner with his learning social environment.
Furthermore, as learning analytics uses indicators for evaluating the progress of learners, bots can be programmed to maximize those indicators and fake the learning behavior!