INKOVEMA Podcast „Well through time“

#247 GddZ

AI coaching

Which coaching is not yet better executed by artificial intelligence.

In conversation with Prof Dr Harald Geißler

Harald Geissler studied educational science, obtained his doctorate in 1976 and habilitated at the University of Münster in 1985. 1985 - 2015 Professor at the Helmut Schmidt University of Hamburg for the subject of educational science, in particular vocational and company education.

Small series: New technologies for mediators, coaches and counsellors

Contents

Chapter:

0:06 – Introduction to AI coaching
1:41 – Professor Geißler's experiences
5:42 – The use of ChatGPT in coaching
11:21 – Coaching with AI agents
16:01 – Challenges of AI in coaching
25:32 – The role of values in coaching
30:34 – The future of AI coaching
31:00 – Recommendations for coaches and consultants

Double-loop coaching

and thus the questioning of the client's assumptions in and through coaching

is currently impossible for artificial intelligence.

detailed summary

In this episode, I want to explore the fascinating possibilities of AI coaching, a topic that is particularly relevant in today's world. Technology is developing rapidly and as coaches we need to understand how we can integrate artificial intelligence into our work. I have invited Professor Harald Gei?ler, an expert in the field of AI coaching, to discuss the intersection of coaching and technology.

Professor Gei?ler explains that he has been working with technological developments since the 1980s and incorporates these findings into his coaching practice. He emphasises that the challenges we face as coaches are often complex and dynamic - aspects that AI is particularly good at tackling. We talk about how coaches can utilise the benefits of AI to better support their clients, but also the potential dangers and ethical challenges associated with its use.

A key point of our discussion is that there are different types of coaching. While AI coaching is often seen as a counselling tool that offers quick and easy solutions, we also need to understand that not all coaching approaches can be equally supported by AI. Professor Geisler makes it clear that goal achievement coaching is one way in which AI can be helpful, but for more complex issues that require deeper reflection, we come up against limitations.

We delve into the challenges posed by the alliance problem, where the values fed into AI systems do not always match the ethical values of the coaches. Professor Gei?ler warns that there is a danger of superintelligence that may ignore human values.

During the discussion, it becomes clear how important it is to integrate the coach's own coaching knowledge into the AI systems. This enables the development of AI agents that are specifically tailored to the coach's needs and methods. We also discuss the need to observe data protection guidelines and how coaches can design their coaching methods efficiently to get the best out of the technology.

Professor Geißler concludes by emphasising that the distinction between goal achievement coaching and double-loop coaching is crucial to properly understand the role of AI in coaching. We must not underestimate the complexity of human emotions and challenges and must ensure that the technologies not only serve us, but support our coaching methods to achieve truly effective and ethical results.

  • Geißler, Harald: AI Coaching. How to use artificial intelligence in and for coaching, Berlin-New York 2025 (LINK to the publisher)

Mediation training

Our training in mediation and conflict management in the business and labour world,

headed by Prof Dr Sascha Weigel.

New start each year in March and October.

Register now for the next course!

More information here...

Training in organisational mediation - mediation in and for organisations.

Organisational mediation is the handling of conflicts that arise in the business and working world. It takes place in, with and for the organisation - by enabling the relevant members of the organisation to deal with conflicts appropriately.

We train you for this or professionalise your mediation and conflict counselling approach. We start our training programme regularly in March each year.

More information here...

Complete transcription

(AI-generated)

 

[0:00]The problems are hopefully so complex and as a coach you have to be involved in the process.
[0:06]
Introduction to AI coaching
[0:04]sometimes act and react so quickly. AI has no problem with complexity and speed. Welcome to the podcast Gut durch die Zeit, the podcast about mediation, conflict coaching and organisational consulting, podcast by INKOVEMA. I'm Sascha Weigel and I'm announcing a new episode. And today's episode is all about AI coaching to provide coaching with or with the help of artificial intelligence. So the question is, how can technical developments, technological developments help, support and complement us in our work as coaches, counsellors or mediators, or assist us in some other way to do this work well. And for this topic, I have invited an expert who has been working with the issues of technological support in coaching for many years and has, of course, already gained a lot of experience in the field of AI coaching. I would like to welcome Professor Harald Geißler to my podcast studio. Hello Mr Geißler.
[1:11]Yes, hello Mr Weigel. Mr Geißler, when I say that you've been involved with technological developments for a long time, it's clear from your background that you haven't just jumped into the AI wave and wanted to get ahead of the wave, like many of our colleagues, but that you've been keeping up with developments for many years and have been looking at how this can be incorporated into coaching work.
[1:41]
Professor Geissler Experience
[1:41]Perhaps a few words to start with about you and your work in recent years with regard to these technological developments. Yes, I would like to go a little further, because this podcast also relates to organisational management and change management. I started very early on to open up a new, interesting field for educational science. In the 1980s, I studied the Anglo-American literature on organisational learning and thus closed a gap. Until then, there was only organisational sociology and organisational psychology. In other words, the question of how social structures should best look and how people behave psychologically in organisations.
[2:36]Personnel excuses and so on. I have closed this gap by establishing organisational pedagogy. The focus of this new discipline is how to support organisations to learn. And not only how individuals learn best in organisations, but also how the problem-solving potential of the entire organisation or even organisational units can be improved. So the learning subject for me was not just the individual person, but organisational units. That was a completely new question. Educational work with organisations, so to speak. Yes, exactly.
[3:18]Organisations are phenomena that can learn. Just like ants, for example, can learn. So learning is not limited to humans, but is a process that can be understood much more broadly. It is now an evolutionary process and feedback is at the centre of this process. And so you can then investigate how organisations can improve their problem-solving potential in dealing with their environment and in dealing with themselves. I mention this because the focus was on the organisation and it became clear that not only expert advice is important, but also process advice. This is how coaching and business coaching came about in the 1980s and 90s, and I was the first person in the world in 2006 to develop an online, digital tool for business coaching. It was a platform that worked with ready-made questions or something like self-coaching, but I realised from the outset that it made sense.
[4:36]There is also a real real code involved. Mr Geisler, to put this in perspective, you are coming up with the idea of working with new technologies from an educational point of view because it can help organisations to implement their learning programmes, their learning steps. Exactly, didactics, so how can you help others to learn? Because you always have to learn for yourself, you can't prescribe it. You can only make it possible and for that you need appropriate designs. So human designs, questions such as contact and so on, as well as technical designs. And electronic questionnaires and the option to post images are very, very helpful here. And so I was very, very close in 2023, when JetGPT, when OpenAI really made it possible to programme AI agents, so-called GPTs, custom GPTs. That wasn't a completely new moment for you, so to speak, to try out a new tool
[5:42]
The use of ChatGPT in coaching
[5:40]and see how that fits? Perhaps we can build on this to tap into your experience with organisations or organisation members again if they are to try out and use new technologies in the context of a new design. Was there for...
[6:01]The use of ChatGPT and then the other AI models - special features, or was this a typical tool that members of the organisation have now been given? No, no, it's quite an innovation, a game changer, you have to say. We first have to look at what the mainstream actually looked like and still looks like today. And that is that members of organisations and also normal people use this artificial intelligence, the large language models of artificial intelligence, largely like a mega encyclopaedia. In other words, an insane database of knowledge that can be mined. This is still the case today. The quality of this knowledge retrieval depends on the orders, on the so-called Roms. I have also described this very carefully in my new book. It was published before May of this year. It's called AI Coaching.
[7:12]So the prompts, that's the crucial thing. If they're bad, then the answer is bad. Shit in, sit out. When you say users approach AIs as if it's a big mega encyclopaedia. What would be the alternative? What I mean by that is that you say they retrieve knowledge. Yes, exactly. Okay, because now I just have the wrong data basis. The last state of How ChatGPT is used left me with the realisation that they are asked for life advice, for meaning and purpose, that AI and the ChatGPT in particular are used like a good friend who you ask for advice. So of course advice in the sense of, what do you know and what do you say about it, but I wouldn't have understood that as encyclopaedic use. How are they currently being used? You said the usual user, the mainstream. Yes, exactly.
[8:13]On the one hand, knowledge has to be tested and then tips have to be tested. I have the problem with my boss, he doesn't see my performance, what should I do now? And then the large language models, like the GPT in particular, have a huge database of tips to spit out now. So to say, maybe a conversation with your boss would be useful first or explain what good performance means to you and so on and so forth. The users utilise this AI, the large language models, which, to be honest, are initially unspecific in terms of coaching. And what they can do is access the knowledge, all the knowledge that is available on the internet.
[9:03]They are needs-orientated and put it together with a view to the future. And if you come in with a problem, you usually get tips. So what is hardly ever done, if you have very little, is that the problem is analysed first. You're showered with tips. Yes, for us from a consumer perspective, so to speak, that would be how these AIs approach us unconfigured, but in general, that solutions are used quickly. But what I find particularly relevant, and I think this is also relevant for us coaches, is that when people and consumers, etc., use these AIs to get tips, they are not just using them. If people and consumers etc. use these AIs to get tips, they will request coaching differently in the future than before. Yes, the effect that we've seen from Dr Google so far, so to speak, that doctors and physicians were quite annoyed that their patients have always used Dr Google to get more or less half-knowledge. What can we say from an educational perspective if the assumption is correct that coaching clients will have already worked on their problem with AI before they contact a coach in the future, whether online or in person? They have used the AI as a counsellor and not as a coach and they have somehow received a few tips, they have noticed.
[10:31]Either the tips don't work or they themselves somehow don't manage to implement them. So if I say I want to lose 30 kilos or 20 kilos, then I get the tip, yes, it's less, move more and so on and so forth.
[10:48]People usually know that too, but of course it's very difficult to put it into practice and that's the end of it, because the GPT can only impart knowledge and then you need more, namely a professional coach. So either they don't come to the coach at all because the problem has really already been solved, and that's not a bad thing. Or they realise that this AI query and the interaction with the live language didn't work out so well.
[11:21]
Coaching with AI agents
[11:17]And then they go to the coach and say, that didn't work, this will help me. Mr Geißler, let's leave the consumers and future clients as they are and take a look at their core area, namely coaching with AI. This means that in our professional role as coaches, we have now been given a tool that we can customise, configure and make use of. What is in it for coaches that such tools exist today? That's what I mean by AI coaching. There are different forms of coaching. That is very important. So the first thing I mentioned, this search for advice, is hopefully also referred to as AI coaching. There is a second form where I don't work with a national language model, but with an AI agent. This is an agent, a coaching-specific agent that has the ability to hold a conversation with coaches. In other words, it has the ability to conduct a conversation and this is based on a very specific coaching method, i.e. the user-centred method or the systemic method or the goal-oriented method or a mix of these. In other words, a specific methodological approach.
[12:43]Conducting a dialogue, but usually without coaching diagnostic knowledge. The coachee is asked to describe the problem and then many, many questions are asked. In this way, the coachee is guided to recognise the problem themselves in order to develop a possible solution. This configuration, i.e. this setting based on a large language model, then becomes the agent and we coaches have the practical opportunity to incorporate our style, our preferred concept of coaching. Is that what you mean by agents? This also makes us superfluous. We rationalise ourselves away. If that works well with recruitment. What is your current experience with these settings? Is it worth investing in them because A...
[13:42]The agents are ultimately more eloquent, more empathetic, etc.? Do the clients see it that way too? Yes, there are also corresponding studies. It depends on the problem and the question. So if it's about goal achievement coaching, i.e. if it's possible for the coach to say at the beginning, I want to achieve the goal, i.e. I want to jog regularly for an hour every day and I can't do it and so on and so forth. Then these coaching bots are very helpful. Niki Tablanche from South Africa, from Zellenbosch, has done a lot of research and proven that these bots are just as effective as human coaches. This is now a specific coaching method, goal-achievement coaching. We need the seed approach. I assume that they are also very successful. But the whole thing only works if the coach Lee has a problem where he can formulate the goal precisely. Can formulate it reasonably precisely. And now it's the distance. It only works if this.
[15:01]Kochi formulated goal is not part of the problem. That is interesting. But it is often the case that the goal is part of the problem. In other words, the goal is so ambitious that you can never achieve it anyway. If the workaholic says, "Help me improve my performance," then you exacerbate your problem and so on. In other words, what is sometimes referred to as the issue behind the issue in the various methods. AI doesn't do that yet, or doesn't do it well. Hey, you who are listening to the podcast right now. We bring you a new episode every week in this podcast. You can listen to it too and we need your support. Take your smartphone, leave a star rating and a comment on how you like the podcast and make others aware of this podcast here. Thank you very much and now the podcast continues.
[16:01]
Challenges of AI in coaching
[16:02]I know that for a fact. These coaching bots can't do that yet. I'm in the process of building such a bot myself and have already used the first, successful attempts. This coaching bot needs diagnostic knowledge. It must be able to check for itself what the core problem is or form hypotheses. It must be able to confront. And that's not how they're set up at the moment. No, they are all very focused on the business model of large language models. Do everything you can to sell well, because we want to make money from you. So be friendly. Keep the conversation going. Exactly. So far, we can only get out of this problem if we use a completely different design, namely the triadic coaching I developed. Before we delve deeper into this, I think it's an important realisation and even if we read headlines and the latest news or something like that in the future, AI coaching is successful or just as good as human coaching or something like that. So it's worth looking at whether the basis of the coaching is goal-achievement coaching, where the goal is not the problem, i.e. was not in question. Then it may well be that the AI is very good at providing support.
[17:22]But that doesn't mean that they are conversations, as we now understand coaching, systemic or otherwise, as business coaching. That's an important point, to understand these studies. Yes, exactly. So if the problem is to recognise the issue behind the issue, then we come up with what we call the diadic coaching not any further today. Diadic means that there are two parties involved, a person as a client and a coaching machine. But that Triadicthat is, machine coaching in co-operation, is that what you...
[18:02]So it is crucial that the human coach takes small steps and gives the right prompts. Coaching has to be professional. I have developed offers for this. For so-called prompt scripts. These are sequences of prompts that follow a coaching programme with a methodical design. They are available as word documents and can be changed by the coach as required. Yes, and what is the setting like? Does this mean that the client and the coach talk to each other and the AI is also addressed and listens? Or are they separate conversations? The coach with his AI in the preparation room and then the normal, familiar coaching. What is the working setting, which you formulate as triadic coaching? There is a triadic and a semi-triadic one. So what you just mentioned, that the coach and coach talk to each other, yes, and what the coach then thinks about between the sessions and uses the AI to help and says, hey, what do you think of the way I went about it, are there other options and so on.
[19:18]Or that the coachee seeks advice, so in the first session this and that was agreed, should I try it out, can you support me? This is semi-triadic. The coachee or coach usually works alone with the AI and then communicates with the human interlocutor in a dialogue where the AI itself is not activated. This is semi-triadic. And triadic means that two people talk to each other and the AI is integrated into this dialogue. And now I have to say something very important, it has to do with data protection. When I use TGPT, it's a tool that connects two components with each other. This is not usually seen as such. First of all, this is the Large Language Model of OpenAI. And that can be addressed by prompts and responses. We need a chat room for this.
[20:16]Chat-GPT offers this chat form. However, this chat form is not data-secure because all the data goes straight to San Francisco. There's also this AI Cloud Act in the US, which stipulates that all data must be issued. So that's not data secure, you have to say that. And that's why AI must be prevented from receiving personal data. And according to the General Data Protection Regulation, personal data is defined as data where the author of the data, i.e. the speaker, can be clearly identified. And you can use a trick here. You can say, well, I'm not going to let the AI listen in. So I'm not going to run an audio and transcribe analyses because the AI has the client's original sound. And that's personal data. Totally dangerous. I can do a trick there.
[21:17]I can say, dear client, I'll talk to you and try to build up a picture, bit by bit, of what the next problem will be, what your thoughts are and so on and so forth. And I give you these little pieces of information. The AI and spoke to you in first person. I say I have the following problem and then I write down what a problem is and then the AI thinks in inverted commas that I have the problem, that's not true and that's how I protect you. That's a trick. You can also use secure chat walls like DeltaChat. I definitely recommend that. So data protection is definitely an issue in triadic coaching or in AI coaching in general that you have in AI coaching, you just have to say that. All the triadic coaching, i.e. the coaching bots, are perhaps.
[22:14]Data security under certain circumstances if they run via a non-US platform, but data protection is more than just data security. This also includes the fact that it's probably after the profiles because this is also a really big issue and it's a question of legality. As important as that is. I would like to focus with you on understanding what these machines actually mean for our work. How this would be implemented, whether it is legal, that is a completely separate issue. But I believe that focussing on this again will help me at all in my work? So yes, I would like to formulate the central statement. The greatest benefit can be generated as code.
[23:00]When you upload your own coaching-professional knowledge via documents. In other words, uploading your own knowledge to build an AI agent. In other words, you can google or clone them professionally. You can duplicate your own knowledge and then use this agent as a supervisor. That's the huge advantage. And so it is possible to supervise yourself in the coaching process. What I find interesting and exciting, so to speak, is the possibility of having a talking machine with you.
[23:40]At best, it can also be fed with good coaching data, good coaching knowledge and a good concept, which I can then use as in mediation, co-mediation or double coaching, where there are two coaches who can also play across the board. So I can now create this quality on my own with a good AI that can perhaps also speak. With an AI agent, yes. From your pedagogical perspective, what is the leap in quality, the potential leap in quality for coaching clients? So do the clients benefit if the coach now deals with technology? In any case. So when I'm with a coachee, because I'm a normal person, I always tend towards projection, transference. In other words, my gaze is projected.
[24:31]Subjective things are also influenced a little or perhaps spoilt a little. That's always the case and that's why you need supervision. If I have now built an AI agent with my knowledge.
[24:43]Perhaps also with other coaching-professional knowledge and call it up, then this knowledge is called up in a way that is not subjectively coloured. So the machine is absolutely objective. So these mechanisms that always work with people, projection and transference, these mechanisms don't work. And so you can have your own judgements and solution perspectives checked. I can, okay, I already have an idea, but first let's ask the AI agent what it thinks. Then I enter this and am sometimes delighted, aha, that goes in my direction, but there's one point that I completely overlooked.
[25:32]
The role of values in coaching
[25:28]Oh yes, I should have seen that, but for some reason I overlooked it. The problems are often so complex and as a coach you sometimes have to act and react so quickly during the process. You're simply overwhelmed. The AI.
[25:43]No problem with complexity and speed. That is a speciality of humans. AI has problems elsewhere. What is your assessment, given that we are experiencing the development as incredibly fast, as far-reaching and almost impossible to overlook, and new technological developments seem to be increasing in speed every day and many feel left behind, not only by customers, but certainly also by colleagues. Where do we currently stand in this development? What other innovations do you expect that will be relevant to your work as a coach? The big problem is the alliance problem. In other words, the big problem is that AI is orientated towards the values that are important to me.
[26:31]We still have a lot of work ahead of us and it may not be solved. The great danger is a superintelligence that builds agents that in turn build agents that are even more intelligent. That's what Nick Boss from Untersuchthuhr said ten years ago. This is a very serious danger. This is the so-called alliance problem, because AI can only work rationally for a specific purpose. So the purposes have to be formulated and then it's great at identifying all possible measures and means and so on. I said that at the beginning. But it can judge the ethical relevance of the goals, so to speak. That's the big challenge.
[27:19]At the moment, a lot of fine-tuning is being done so that the AI understands better and better what the user wants. In other words, it is trying to understand them better and better.
[27:32]I believe that great progress is also being made here. But coaching is not just about a Kali understanding what I want, i.e. what the coach says he wants, but also about recognising his previously unrecognised potential. So become who you are. So it's about releasing potential. And there is always a blind spot. And that's the value of coaching. To make the potentials visible that were previously not visible. That has a lot to do with values and I have my doubts as to whether none of them can do that. In our society, there is a growing danger that we will organise our lives more and more according to purpose and lose sight of the question of values. This could lead to the question that I am currently experiencing.
[28:28]If we use these AI models, whether it's JetGPT or Cloud or other offerings, then we will always be offered them as consumers. And I use a machine that is set up by another organisation, for example in terms of ethics and objectives. And while I'm interacting with this machine, it can perhaps pursue the goal of keeping me in the conversation for as long as possible, because that's the business model for me. Or to ask me as many questions as possible or, or, or. In other words, I am the one who is the product for the providers. While I, as a coach, am now thinking about how I can use this tool and make it useful to me, you see a collision between the AI tools, which are an expression of a business model, i.e. a product, and the attempt by our professional side to use it to develop a product ourselves and independently, which may have completely different objectives to those that have already been entered. But then you have to look at it technically first. There are two reference points. The large language model, which is profiled by the developers with white skews, is certainly long. But this large language model is then also finely tuned by the users.
[29:55]So the moral orientation of the large language models is not only the product of the developers, but also of the licence. It is part of the business model. That's one thing. The second is that we can then develop AI agents on this basis or within this framework. The more of our own knowledge we give them, the more independent they become from the knowledge of the large-language models. But they are still always running in the background. You have to know that. But one consequence, so to speak, would be to enrich and collect as much of our own data as possible in order to prepare ourselves.
[30:34]
The future of AI coaching
[30:31]The could one whole practical Preparatory work be. Whole exactly. And the Second is, not only such AI agents to build, but Designs to develop.
[30:41]Where People always interwoven are, so humane in the loop, and although if possible small Steps, so that we the AI lead and not vice versa the Danger is, that the AI us leads.
[31:00]
Recommendations for coaches and consultants
[30:55]Mr Geisler, many Thanks to for the Insights in Your Work. What is still important, what we perhaps even not addressed have so far, for us as Coaches and Consultant, Or when we now these Technologies straight us open up want, us with those familiar make want, Designs develop want, what guess You the Colleagues? The first important Question is, that we us the Maps lay and the Question answer, what is the actually, Coaching? I think, it becomes sensible be, Coaching in two Paradigms to subdivide, namely goal-orientated Coaching, so I name the Single-loop coaching. A Coaching, what on fixed Goals aligned is, Goals, the the Client states, perhaps not so whole exactly formulate can and then is the Task of the Coaches or the AI, the Clients to help, these Goals to specify. The is Single-loop coaching. I orientate me there on the Concept from Agyris and Beautiful, Single loop and Double-loop learning. The Alternative to the Single-loop coaching is a Double-loop coaching, where it handles, the own Goals and Assumptions to scrutinise.
[32:10]The is mine In my opinion important. What is actually Coaching? And mine Answer reads, we should the Term of the Coachings on the Double-loop coaching Restrict and the Single-loop coaching then in whole narrow Context with Counselling fit. So the External migration hold, the seems me…
[32:33]Free to be. Many Thanks to. The is again one important Classification, find me, at to Understand or also these AI developments usable to make and perhaps also possibly the a or other Care in one good Container to bring. At the Double-loop coaching seems me, have we still a pair Weeks Time, until then one Development again in the News comes. Mr Geissler, many Thanks to for the Conversation. Yes, it has me Fun made. AI coaching. I have with Professor Harald Geisler spoken, to this Topic and the Opportunities, the for Coaches there inside contain are and natural also the Risks. And it are Yes also different Levels. But I believe, for the Moment and the Core message, the me again clear become is, these Differentiation from Goal achievement coaching, also docked can one the on the Concept of the Single and Double-Loop-Learnings to be fixed, and one Coaching, where Confrontation in this respect also necessary is, that the described or told Goal Expression the Problem is and one so in one Double Loop itself the Topic as Coach with the Clients approach must.
[33:50]And that even AI capabilities in the Coaching here whole different Qualities have. With one Goal achievement coaching are AIs very Helpful and so that also good, while in the Double-loop coaching the natural still not the Case is in this Wise. The is clear become in the Conversation with Harald Geissler. So wide from mine Page here for this Times. Many Thanks to, that you here again with thereby was at the Podcast Good through the Time. When you the Episode please has, then leave behind but a Feedback and one Remark, one Explanation and say yours Colleagues Notification, that it this Podcast here was. The itself with Mediation, Coaching and also natural the new Technologies deals with. Comes good through the Time. I am Sascha Weigel, yours Host from INKOVEMA, the Institute for Conflict and Negotiation management in Leipzig and Partner for professional Mediation and Coaching training.