AI is changing the world

AI is changing the world. Maybe, experiencing these changes from the inside, as citizens of the game changer superpowers, we don’t really realize the violent impact of disruption on our daily life. The violent impact will come when, while cooking dinner for our beloved family we realize that an Artificial Intelligence, linked to the watch of our children, through monitoring pulse rate, body heat and oxygen in the blood, will be able to recommend the best cartoon for them to watch or the best vegetable to eat. And just at that point we will really think “ok, what’s happening here?”. But for now, seeing these changes in “rallenty”, reading about one “small” innovation at a time (just because we can’t read simultaneously more than one word at a time), doesn’t really seem to shake us. Well, long story short, the world (and the business world firstly) is changing and it’s changing fast.

When speaking about innovative practices, people can lag behind, Businesses can’t

We, as people, can surely remain back a few steps without risking bankruptcy. Businesses can’t. Nevertheless, when the businesses on which we focus are consultancy societies that, in order to survive, formulate strategies to make other societies stand out, within a sea of competitive societies backed up by other societies. In this mess,it is inevitable that consultancy firms must not only keep up with innovation, but even direct and guide it. Regarding this last statement I’m pretty sure they will. How? Well for two main reasons, they are constantly immersed in a network of stakeholders (clients and partners) who are informed or need to be informed regarding cutting-edge technologies and trends and because of their human-centered nature (with relatively low fixed assets) that allows them to implement new methodologies or sell new products in a fast and nimble way. Now, when we speak of implementing new methodologies and selling new products, what are we referring to? Well, in this article, we are referring to implementing and selling Artificial Intelligence languages.

How can new languages be created? A chat with Carlo

In order to understand how new languages are created, I interviewed my long-time friend Carlo, Master of Science in Machine Learning and Research Engineer at CISPA Helmholtz Center for Information Security. In simple words Carlos’s daily job involves adjusting leaks and improving the state of the art of AI and deep learning models. I interviewed Carlo in order to understand what it means for Consultancy Firms to create their own AI language and what are the variables involved. Speaking with Carlo I understood that there are two main protagonists in this process, the data (the oil) and the model (the engine). While the parameters of the models can be obtained and used by all, since societies such as Meta make them public and useable (we are speaking of pre-trained, large and generalized language models), the data needed to make the model speak the “consultancy language” (through the “fine-tuning” process) is more difficult to be found. For this reason it is more likely that big companies such as PwC, EY, KPMG, that already have tons of terabytes of in-house data can more easily get the engine to work. Small companies instead, are obligated to download it from public online sources such as Kaggle or Wiki English that still provide great quantities of good quality data. This doesn’t mean that small companies can’t do it, it just means that for them it’s probably going to be more costly. So once the consultancy company in object has on the one hand Meta’s pre-trained AI model and, on the other hand, the data, Carlo takes the field. His role is to use the data collected making the model speak the strategic consulting or marketing consulting or HR consulting language (depending on the data collected), delivering a product that can increase the efficiency of work within these companies. And how does he do this? Through the fine-tuning; a process that consists in training the pre-trained, large and generalized language model and making it better suited for a particular application.

Deep diving the apparently impossible process

This process involves the use of more than 20 GPUs (Graphic Processing Units, the billions of operations needed to train these models are performed on specialized computers built for quick and parallel processing). Once the model has been trained to serve a particular mansion or “speak” a particular language, the model can be prompted and used to generate new text or voice. What happens in this second step (called inference or prediction) is often referred to as a black box, since the human mind can’t really understand what is going on under the hood and why exactly the model is generating a single specific output (but this is another story). So once the fine tuned model is created, it needs to be put in production. Putting a fine-tuned model into production is a complex process. Through CI/CD pipelines, which means Continuous Improvement and Continuous Development, the model is updated and trained with data and questions, and every time a new output is generated, feedback on that output is feeded back to the model, to ensure the full lifecycle of the CI/CD pipeline. This helps ensure the model works effectively and doesn’t give unwanted answers (such as racist answers or information on how to conduct illegal activities). Finally, after completing this phase, the model is ready to be used by the consultancy firms (while needing continuous training with new and updated data).

The outputs of the impossible process, the case of Lilli

This process, which seems long and expensive, is actually doable, thanks to the work of specialized engineers and consultants, such as Carlo, that concludes the interview making his own prediction: “I think the major consultancy companies, in five years, will all have ad hoc fine tuned AI models integrated into their systems. Their employees will use them to boost their performance working in a more efficient and data driven environment.” And someone already started! McKinsey, this year launched “Lilli,” an own generative AI solution that aggregates all Mckinsey’s knowledge and capabilities in one place. The data used to train Lilli relies on more than 100.000 documents and interview transcripts that the firm collected during years of hard work. For Adi Pradhan, an associate partner who specializes in technology strategy and transformations, Lilli is “a thought-sparring partner” ahead of meetings and presentations. He uses Lilli to look for weaknesses in arguments and anticipate questions that may arise, to tutor himself on new topics and make connections between different areas on the projects.

Acknowledgement Statement

This blog is part of the student writing competition in Management Consulting Master Program at the School of Business and Economics.

References

• How Professionals Adapt to Artificial Intelligence: The Role of Intertwined Boundary Work, Journal of Management, 2023
  [https://onlinelibrary.wiley.com/doi/epdf/10.1111/joms.12936]
• How professional services firms dodged disruption, MIT Sloan Management Review, 2023
  [https://sloanreview.mit.edu/article/how-professional-services-firms-dodged-disruption/]
• Consultants Emerge as Early Winners in Generative AI Boom, The Wall Street Journal, 2023
  [https://www.wsj.com/articles/consultants-emerge-as-early-winners-in-generative-ai-boom-8df71d38]
• Killing Strategy: The Disruption Of Management Consulting, CBINSIGHTS, 2020
  [https://www.cbinsights.com/research/disrupting-management-consulting/ ]
• Revolutionizing Decision-Making: How AI and Data Analytics are Transforming Management Consulting Practices, Medium.com, 2023
  [https://medium.com/accelerated-analyst/revolutionizing-decision-making-how-ai-and-data-analytics-are-transforming-management-consulting-94b98a8b1e1b]
• Meet Lilli, our generative AI tool that’s a researcher, a time saver, and an inspiration – McKinsey Blog, 2023
  [https://www.mckinsey.com/about-us/new-at-mckinsey-blog/meet-lilli-our-generative-ai-tool]