Brave New World. Full of Intelligent Agents.

Subscribe to our Newsletter on LinkedIn to stay tuned

“Brave New World. Full of Intelligent Agents. Their existence was stitched seamlessly into the fabric of everyday life, their luminous eyes and sleek synthetic skins blending into crowds of humans with deceptive ease. These agents, designed for empathy and engineered for efficiency, navigated the sprawling urban landscapes with a grace that belied their artificial nature. As the sun dipped below the horizon, casting a copper glow over the city, they moved among us, their algorithms quietly shaping the course of human interaction, their presence both a promise of a better future and a reminder of the complexities introduced by their creation.” - ChatGPT 4.0

prompt: “Act as a professional science fiction author. Write me one paragraph starting "Brave New World. Full of Intelligent Agents.”

In this Future Insight, I would like to share my thoughts about the development of “intelligent” AI agents, from the first seat row. Promise, I will not go gentle into this good night.

The year 2024 will probably be proclaimed as the year of AI Agents—at least for the first half of the year. But I’m afraid the following quarters will be filled with disappointment.

Here are a few important points to consider:

Startups built around ChatGPT 4.0 are likely to be significantly impacted by the release of ChatGPT 5.0, a development confirmed by Sam Altman. Same applies for any business build around specific LLM model.
Consultancy firms that are repackaging their old machine learning projects as modern AI solutions may find themselves losing credibility and clients.
Professionals who were misled by AI “charlatans”- oh sorry- promoters might face difficulties in rebuilding trust with legitimate AI technologies.
The car design industry will resist change for as long as possible, but eventually will need to adapt. Those who don’t want to adapt probably should order a copy of “On The Origin of Species” already today.
Meanwhile, the rest of the world will continue to enhance their lifestyles using general-purpose AI systems.
Agent-Based AI, developed by specialists in their fields, will likely prosper and excel.

Let me dive into the basics to explain further.

Foundation Models.

The LLM revolution have been gaining visibility last year with the general-purpose foundation models based on large-scale pre-training on broad-scale internet data from a variety of sources. These are GPT-series, LLaMa, Claude, and others. Pre-trained models have a fixed set of knowledge and understand general concepts.

What does this mean? For instance, if you ask the model to provide an opinion or summary about the projected growth of the BEV (Battery Electric Vehicle) market, you will likely receive a fairly accurate answer. Because it has been a hot topic recently. However, if you ask a more specific question, such as 'What are the most important features of light bulbs in Lithuania?' you will not get a precise response.

Agent-Based AI

Agent-based AI is different from traditional AI because it can create flexible behaviors that respond to changes in the environment. What’s crucial is that AI agents are becoming specialized in specific areas of expertise. Once we successfully develop specialized AI agents, like Intended Future's agent focused on the Perceived Quality of Cars, the next natural progression will be to develop multi-modal AI products.

Who can do it?

There is a well-known gap between academic knowledge and industrial practice. Academia tends to focus on long-term vision but often lacks practical, hands-on knowledge about the subject matter. On the other hand, the Industry is always short-term actions because the Industry has to survive. As simple as that.

To connect these two worlds, we need explicit applied academic knowledge and approach to frame the unknown quickly. On the other side, we need all the power and infrastructure of Industry to implement new knowledge in practice swiftly.

What does it mean for the design?

Archimedes famously said, “Give me a firm place to stand and a lever, and I can move the Earth.”

I would adapt this to our context by saying, “Give me a comprehensive framework describing the principles of car design, along with Vizcom (or a human designer), and I will create an AI agent capable of designing cars.”

This raises a question: Do we still need humans?

I've been wrestling with this dilemma since last November. In the Perceived Quality (PQ) domain, we have developed the most extensive framework, which was built over four years with unique insights from various automotive OEMs. We have made the Perceived Quality Framework (PQF) publicly available to give back the knowledge shared by car designers and engineers. However, PQF is not static; we continuously fine-tune and refine our models.

We own extensive data on what car drivers around the world value most. We are capable of performing detailed craftsmanship assessments of complete vehicles in-house and training our models based on these evaluations.

Do we need human designers and engineers?

Let me keep you intrigued. I will answer this question in the next edition of Future Insight.

However, I'll give you a hint about the main theme of the article:

We're not discussing the survival of the best “design species”; rather, we're talking about the “fittest.”

References:

Durante, Zane, Bidipta Sarkar, Ran Gong, Rohan Taori, Yusuke Noda, Paul Tang, Ehsan Adeli et al. "An interactive agent foundation model." arXiv preprint arXiv:2402.05929 (2024).

Wang, Yizhong, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A. Smith, Daniel Khashabi, and Hannaneh Hajishirzi. "Self-instruct: Aligning language models with self-generated instructions." arXiv preprint arXiv:2212.10560 (2022).

Durante, Zane, Qiuyuan Huang, Naoki Wake, Ran Gong, Jae Sung Park, Bidipta Sarkar, Rohan Taori et al. "Agent ai: Surveying the horizons of multimodal interaction." arXiv preprint arXiv:2401.03568 (2024).