Are the AI tutors from Open AI, Google and Anthropic effective?

 One of the major applications of Gen AI in education, is to support personalised education in the form of individualised learning support for each learner. However, due to the way in which AI is structured, its propensity to 'hallucinate' is still a challenge. Additionally in our work with chatbots (see project report for details) the chatbot has a propensity to 'drift' from its training after several students have used it. Therefore, there is still a challenge when using AI to underpin personalised learning.

This mashable.com article runs a test through the AI tutors provided by OpenAI, Google and Anthropic. The article provides for an interesting overview, the process for testing and then an summary of the performance of each. Links for each of the summaries bring you to a more detailed write up. 

The findings/discussion will not be new to teachers. Importantly, the writer decides that chatbots cannot replace great teachers; reinforcing that learning is social. 

Another post from AI toolkit undertakes a similar discussion. Although personalised learning is a 'holy grail' in education, the ways AI tutors go about provisioning answers, rather than supporting, nudging, encouraging learning is its disadvantage. When learners do not have to put effort into the learning process, learning does not take place. They end up with a 'cognitive debt' because the brain is not engaged in 'solving the puzzle' and 'making meaning'. In turn, learners develop 'metacognitive laziness'  and depend on AI to do the hard work of thinking required to increase learning and develop expertise.

Of note is the status of AI in education. This post in Education Futures warns of the ways education is losing control as commercial interests become the providers of learning platforms. Although written within the US of A context, many of the arguments laid out are universal and global. AI is not only a form of technology, but can dictate what is learnt, what is accessed (or not) and removes teacher and learner agency. 

Therefore, we need to be careful when thinking about AI-supported personalised learning. The design of learning into the AI agent/chatbot is essential to ensure that AI helps learning to take place and not replace learning.

Eportfolios and AI tools: webinar notes

 Attended the Eportfolio workshop - how to get started with AI tools for portfolios. The session was delivered both online and f2f (La Trobe- Melbourne) and organised by ePortfolios Australia and ePortfolios Ireland. 

The session was presented by Dr. Megan Mize who is Director of ePortfolios and Digital Initiatives in the Digital Initiatives Success Centre at Old Dominion University. 

Alison Miller welcomes everyone and acknowledges Country and the various organisations who have contributed towards this webinar. Helen Chen introduces AAEEBL - Association for authenic, experiential and evidence-based learning. Megan begins with overviewing the AI tools that are already part of eportfolio practice. Slide deck of presentations are found at google docs. 

AI's current role could be to support grammar/syntax, accessibility, colour design and content generation. All of these have been available even before advent of ChatGPT. Introduces ideas of how to leverage Gen AI for faculty to provide more individualised support to learners. Proposed the Riff chatbot as a tool which has potential (as discussed with Leticia Britos Cavagnaro).

Riff chatbot can be useful to help learners how to practice reflection. A demo /try out of Riff chatbot followed for participants to evaluate the tool. The chatbot provides some good prompts on statements which were presented to it - a bit similar to 'ask watson' it repeats back what you contribute but pushes the user to 'dig deeper'. Riff chatbot has option of linking with Moodle. Check this blog for using chatbot to support reflective activity and up-coming book by Cavagnaro.

Then provided an opportunity to try another chatbot - Poe chat. Difficulties getting in through institutional and gmail :( (but did arrive in my institutional after about 10 minutes). Demo undertaken to show capabilities of the chatbot. Prompt engineering is important. Tried the 'colour my site' chat which was set up to provide more targeted information to users. Demonstrated how to set up the chatbot to undertake more targeted responses. An authentic activity not only for teachers but also across many occupations. For example, setting up a bot for a business to answer client enquiries etc. Testing the bot is important to see how it responses to a range of inputs. 

Then provided opportunity to 'build a bot' on Poe. Straight-forward process. Also accessible as IOS or Android app. Poe uses ChatGPT and Claude, so not sure how different the outputs are to just using ChatGPT. Claude has a better privacy policy when compared to many other Gen AI tools. 

Discussed the challenges and the need to test extensively, ensure learners triangulate the outputs, and be aware that 'input = output' holds, and the privacy and ethical issues around using the chatbots. Possibilities for learner driven personalised learning are many. 

Then discussed the ethical concerns with regards to AI and eportfolios, summarised in this blog. Q & A followed with most focusing on practicalities and logistics of using chatbots to support personalised learning.

A good session, providing good examples of how to better use AI to support learning, especially for reflective learning.