Why AI Matters for Businesses Today
The broader picture: Levels of machine intelligence and the path to AGI
Why AI Matters for Businesses Today
We don’t start from scratch – AI’s role in enterprise value creation
Navigating the change – Recommendations for the way forward
Elevate Your Business with AdEx Partners’ AI Expertise
Key Takeaways
AI plays a pervasive role in daily life, specializing in distinct functions that enhance decision-making and efficiency.
Generative AI is pushing the frontier of automation, yet its current impact on enterprise value generation remains limited.
Organizations must integrate Specialized and Generative AI to leverage data-driven decisions and innovative content creation effectively.
Artificial intelligence is omnipresent in our private and professional life, from speech recognition in our smartphones, to personalized recommendations in music streaming and customer-centric shopping experiences. Most of today’s AI applications operate behind the scenes, unseen by the end user, and are highly specialized for particular functions. We call these AI applications Specialized AI to emphasize the focus of traditional “Discriminative AI” on specific tasks and capabilities to support decision-making. With the rise of Generative AI, new applications like ChatGPT, Copilot, and Midjourney stand out for their broad applicability and ease of use, seen through their high adoption and popularity. Instead of focusing on a specific application, these general-purpose tools can be applied to a variety of problems. In addition, the initial hurdles for utilization are very low: instead of programming code, they can be controlled using natural language, no complex preparations such as data collection, training, model tuning are required, and the outputs are available within seconds.
While Generative AI is pushing the frontier of automation as the evolving boundary between tasks that can be performed by machines and those that still require human intervention, its impact on enterprise value generation is still limited. It could be argued that in 2024 generative AI is capturing 90% of the attention in the AI space yet adding none or limited net value to organizations. Despite GenAI’s impressive potential, Specialized AI has been and will remain integral to generating value within organizations.
The emerging necessity for organizations is to integrate Specialized and Generative AI, fostering an environment where data-driven decisions are complemented by innovative content creation. This integration aligns with the traditional logical-rational frameworks of companies, which have historically favored targeted analytical methods grounded in data. The challenge and opportunity now lie in harnessing both approaches to enhance efficiency, foster innovation, and bolster competitiveness in the dynamic business arena.
The broader picture: Levels of machine intelligence and the path to AGI
Key Takeaways
Understanding machine intelligence levels is essential as we approach Artificial General Intelligence (AGI).
Task complexity influences the selection of AI approaches based on adaptability, collaboration and skill requirements.
The spectrum of machine intelligence ranges from task-specific AI to the goal of Omnipotent AGI.
IWith Artificial General Intelligence (AGI) on the horizon, a better understanding of the term AI is essential to accurately spot the applications for Specialized and Generative AI models available today while being prepared for further technological development. AGI, which represents AI with capabilities to rival or exceed those of a human, remains partly theoretical at this stage but holds the potential to replicate human-like cognitive abilities, including reasoning, problem-solving, perception, learning, and language comprehension. Large Language Models like OpenAI’s o1, released in September 2024, already exhibit strong reasoning capabilities and complex thought processes, achieving top results in scientific benchmarks and surpassing human PhD-level accuracy in select tests. Despite the lack of standardized criteria for defining AGI, various test procedures have been proposed over the years including the well-known Turing test from the 1950s, the Ability to Learn New Tasks, and the Coffee Test. Depending on the definition adopted, estimates for when AGI might be achieved vary significantly, ranging from a few years to several decades. This is particularly relevant in light of advancements in modern AI systems, such as the aforementioned GPT-4 model family, which exhibit significant progress in multimodality by integrating vision, speech, and text capabilities. As research continues to evolve, understanding these developments is crucial for anticipating the future trajectory of AI technology.
To guide the decision-making process in selecting AI approaches, examining the complexity of the tasks at hand proves to be a valuable starting point. Task Complexity is a multifaceted concept to measures the intricacy of tasks by intellectual efforts (like problem-solving and creativity), technical skill requirements, and the extent of variability and adaptability needed as tasks evolve. It also considers the social and collaborative demands, including the necessity for communication, coordination, and teamwork, especially in dynamic and situational contexts. Task complexity increases when tasks become more dynamic, require significant adaptability or extensive interaction and collaboration. The following model structures this view into three dimensions.
The Task Nature
The Task Nature dimension categorizes tasks by their predictability, with repetitive tasks at one end, characterized by predictable actions and minimal decision-making, like assembly line work. At the other end, dynamic tasks demand continuous adaptation and address novel scenarios, such as in strategic planning and creative projects. Scope delineates the range of knowledge and skills needed for tasks, with specific scope tasks requiring deep, domain-specific expertise, like tax accounting, and general scope tasks demanding a wider skill set across domains, such as in agile project management. The Autonomy of Interaction dimension evaluates collaboration necessity, distinguishing between tasks that are independent, like individual software code contribution, and those requiring teamwork, such as large-scale software development.
This model helps to classify different levels of intelligence and characterizes AI approaches. Artificial General Intelligence embodies the ability to understand, learn, and apply knowledge across an unlimited range of tasks, mirroring the comprehensive and adaptable intelligence of humans. This contrasts with narrow AI, which is limited to specific tasks or domains. AGI sets a trajectory towards the ambitious vision of AI surpassing human capabilities across all tasks. The current AI landscape reveals a spectrum of machine intelligence levels, each defined by its task complexity capabilities, highlighting a stepwise approach to achieving Omnipotent AGI that transcends human abilities and underscores the evolving nature of AI:
Task-Specific: Performs a specific task at human-competitive or economically sufficient levels without broader applicability.
High-Performance Task: Executes a specific task with superior efficiency, speed, and safety, offering significant economic benefits over human performance.
Domain-Focused: Handles a range of tasks within a specific domain with adaptability and learning but lacks skills outside that domain.
Multi-Domain: Shows adaptability and problem-solving across multiple unrelated domains, demonstrating broad but not complete general intelligence.
Standard AGI: Performs a wide array of tasks at a level comparable to an average human, showing general intelligence and adaptability.
The classification of machine intelligence levels reveals a spectrum of AI capabilities as shown in the table below. Current AI technologies predominantly fall in the “domain-focused AI” or “multi-domain AI” levels. Recent advancements have shown a significant push towards enhancing the capabilities of multi-domain AI, indicating a trend towards more versatile and broadly applicable AI systems.
| Scope | Task Nature | Autonomy of Interaction | |
| Task-Specific AI | Specific | Repetitive | Independent |
| High-Performance Task AI | Specialized | Routine | Minimally Interactive |
| Domain-Focused AI | Mixed | Structured | Semi-dependent |
| Multi-Domain AI | Broadened | Semi-Dynamic | Moderately Interactive |
| Standard AGI | General | Dynamic | Collaborative |
| Peak AGI | Cross-Disciplinary | Highly Variable | Highly Interactive |
| Artificial Superintelligence | Every scope | Every task | Fully collaborative |
Why AI Matters for Businesses Today
Key Takeaways
The analogy between AI and the human brain reveals how Specialized and Generative AI serve different functions
Specialized AI excels in analyzing and classifying data, while Generative AI is adept at generating novel data and insights
Understanding the strengths and limitations of each AI type is crucial for effective application in business contexts
One of the most popular analogies to illustrate the complex mechanisms of AI systems is the human brain, which served as inspiration for the development of artificial neural networks. The analogy between AI and the human brain is not only illustrative, but also functionally revealing.
Different regions of the human brain are specialized for distinct functions, with each area handling particular tasks. This lateralization of brain function refers to the tendency for certain neural functions or cognitive processes to occur predominantly in either the left or right hemisphere of the brain. For most people, the left hemisphere of the brain is primarily responsible for language processing as well as mathematical skills and operates strongly analytical and logical. In contrast, the right hemisphere is dominant in spatial perception and visual recognition and, thus, is often also associated with creative processes and intuitive performance. To solve cognitively demanding problems, a combination of these distinctive functionalities is often necessary.
Similar to the lateralization of functions in the human brain, Specialized and Generative AI approaches differ significantly in terms of the range of tasks that can be performed and their associated limitations.
Specialized AI
Specialized AI models aim to recognize distinguishing features in large data sets. They are trained to analyze and classify data and make predictions by identifying characteristic properties to discriminate between different groups within the data. These models are particularly useful for applications that require accurate separation or categorization of large-scale data, such as recognizing objects in images or generating next-best offer suggestions.
Generative AI
Generative AI models are designed to generate new data that resembles the patterns of existing data sets. They can be used to generate new content, such as text, images, videos, audio or code. These models learn the underlying data distributions and can generate new instances that are similar in structure to the original data. They are suitable, for example, for the generation and translation of texts, the creation of photorealistic images or the design of molecular structures for new drugs.
For businesses it is important to understand the pros & cons of these concepts to fully leverage the capabilities of the differing AI technologies. Driven by the recent impressive advancements in Generative AI, this technology has gained preference for the default AI solution for all sort of challenges including image understanding, code refactoring, or text classification. Determining the tasks that Generative AI actually can or can’t perform well is often challenging. The following framework provides selection criteria to guide the selection of suitable AI approaches:
Precision: Consider the significance of a
model’s output accuracy or its margin for error. Is there room for slight inaccuracies, or is it permissible to make small manual corrections to the
outcomes?
Robustness: Consider how well the model can
respond to irregularities in the data. Can the model maintain its performance with noisy data or do these lead to unwanted hallucinations? How does the model cope with unexpected information or conditions that it did not experience during its training?
Interpretability: Determine how important a comprehensible decision-making process of the AI model is. How easy is it to understand or explain the reasons for different outcomes, especially if the model is able to generate a range of outcomes from the same input data?
Consistency: Assess the importance of the model
delivering uniform outcomes when presented with identical inputs. Is predictability in results a necessity, or is a degree of variation in outcomes tolerable?
| Specialized AI | Generative AI | |
| Precision | Optimized for high precision in specific tasks with consistent data Capable of learning complex structures from labeled data for accurate predictionsIdeal for applications where exactness is critical | Versatile in adapting to a wide array of dataTrained to generate novel data and insights May prioritize creativity over precision in certain contexts |
| Robustness | Strong performance on clean, well-defined dataReliable at handling noise and anomaliesConsistent behavior in expected operational environments | May produce random results with noiseLess robust to unseen patternsSusceptible to hallucinations |
| Interpretability | Predictable outputs for given inputsTraceable decision logic (depending on algorithm)Consistent decision-making process | Diverse and rich outputsSemantic similarity in varying outputsEasy access to results through natural language interaction |
| Consistency | Identical inputs yield identical outputsHigh reliability and trustworthinessEssential for tasks requiring uniformity | Potential for output variability for same inputsFlexible in content generation (e.g. format, style)Adjustable for creative or diverse solutions |
“Generative AI is the Swiss Army knife of tech tools, versatile for those unpredictable business scenarios where adaptability is key. However, when the task at hand is as specific as cutting a tree, the precision of specialized tools like an axe or chainsaw reflects the domain where Analytical AI excels.”
We don’t start from scratch – AI’s role in enterprise value creation
Key Takeaways
Understanding machine intelligence levels is essential as we approach Artificial General Intelligence (AGI)
Task complexity influences the selection of AI approaches based on adaptability, collaboration and skill requirements
The spectrum of machine intelligence ranges from task-specific AI to the goal of Omnipotent AGI
To thoroughly understand the data analysis and AI enterprise landscape, it’s essential to categorize AI approaches within existing frameworks. “Foundational Data Analytics” lays the groundwork essential for elementary analysis and preparation of data. “Specialized AI” advances into the realm of predictive analytics and decision support, managing more intricate data configurations. “Generative AI” is at the forefront, dedicated to generating novel data and content, thus extending AI’s reach from analyzing and interpreting data for decision automation to areas of creating unique content and ideas.
Foundational Data Analytics
Characteristics
Basic statistical techniques and foundational data analysis tools including descriptive statistics, aggregation, and basic correlation on analysis
Purpose
Explore and understand data, summarizing patterns and preparing datasets for advanced analysis
Limitations
Confined to examining current data without engaging in intricate predictive modeling or the creation of new data and avoiding inferences about causation or future trends
Specialized AI
Characteristics
Basic statistical techniques and foundational data analysis tools including descriptive statistics, aggregation, and basic correlation on analysis.
Purpose
Explore and understand data, summarizing patterns and preparing datasets for advanced analysis.
Limitations
Confined to examining current data without engaging in intricate predictive modeling or the creation of new data and avoiding inferences about causation or future trends.
Generative AI
Characteristics
Basic statistical techniques and foundational data analysis tools including descriptive statistics, aggregation, and basic correlation on analysis.
Purpose
Explore and understand data, summarizing patterns and preparing datasets for advanced analysis.
Limitations
Confined to examining current data without engaging in intricate predictive modeling or the creation of new data and avoiding inferences about causation or future trends.
The differences between the approaches can be highlighted via their input data, computational logic and the results from applying the logic to the inputs. Input indicates whether the structure of the data is known. Computational logic refers to the comprehensibility and interpretability of the calculations performed. Results describe whether the result space of possible solutions is clearly delimited and defined. Note: These statements refer to how the approaches are applied to processing new data, not the machine learning training phase, for example.
| Approach | Knowledge source | Input data | Computational logic | Results | Insight & Data Flow |
| Foundational Data Analytics | Business logic derived from human expertise | Known | Known | Unknown | Structured input data is analyzed with clear logic to gain new insights. |
| Specialized AI | Data-driven by supervised or unsupervised learning | Known | Unknown | Unknown | Machine learning algorithms automate decisions by processing new structured data within a defined result space for insights. |
| Generative AI | Self-supervised learning & instruction fine-tuning | Unknown | Unknown | Unknown | New, non-deterministic outputs are generated from inputs using learned structures. |
Historically, companies have concentrated on Foundational Data Analytics and well-delineated Specialized AI applications for data-driven value generation and automation. High levels in the the PRIC criteria spectrum are particularly desirable in these contexts. Meanwhile, tasks requiring creativity and complex cognitive abilities have remained beyond the reach of automation. With the advent of Generative AI, the frontier of automation is shifting, redefining what can be automated and thus broadening the spectrum of tasks that can be automated using AI.
Navigating the change – Recommendations for the way forward
Key Takeaways
Organizations should adopt a balanced AI strategy, recognizing the strengths of both Generative AI and Specialized AI based on specific use cases
Staying informed about the rapidly evolving AI landscape is crucial for making effective technology decisions and optimizing model performance
A strategic approach to model selection is essential, as the market for Generative AI will diversify, requiring careful consideration of speed, cost, and quality
As with any disruptive development, the initial exponential increase in performance is difficult to understand and extrapolate into the future. Companies are faced with the challenge of evaluating these developments for themselves and finding a suitable plan that is flexible for the future.
The following recommendations for action will help in navigating these stormy waters:
The Generative AI hype surrounding ChatGPT and co. tempts organizations to focus their AI strategy entirely on these models. Despite impressive performance, however, Specialized AI models continue to achieve better results in many areas and are superior for the evaluation of structured business data, numerical mass data such as sensor measurements or complex mathematical optimization. In choosing the right technology for the task, it is important to consider the strengths & weaknesses each have. Which AI to use starts with the use case, not the AI model.
Due to the fast pace of the AI market, it is difficult to make a technology decision in favor of a specific model. Possible options include the use of standard models (e.g. GPT-4, LLaMA 3), performance optimization by means of prompt engineering or fine-tuning as well as training your own models from scratch. In the foreseeable future, the quality of results from standard models will continue to increase due to mounting capacities of processable data (context sizes) and the more efficient design of model parameters. This development will most likely reach a plateau in the medium term, where general output quality can no longer be significantly increased. At this point, there is a great opportunity for companies to achieve specific performance gains through their own model optimizations, be that through the addition of company-specific data or other means of model refinement.
The differentiation of the market for Generative AI models will lead to the availability of models in different flavors. Differentiators will lie in particular in the speed, cost and quality of results generation. The ChatGPT moment of “one model fits all” will not last. As AI model families evolve for specific tasks, AI application becomes more nuanced – An example is OpenAI’s GPT-4 family, differentiated into GPT-4o for high-intelligence, multi-step tasks and GPT-4o mini for fast, lightweight tasks prioritizing cost efficiency. With o1 and its mini variant, a new family of models arises for performing complex reasoning. Aclear prioritization of use cases and selection of suitable models appears key to efficient model use and the creation of strategic competitive advantages. As the market matures, we can expect a proliferation of specialized models tailored to distinct industry needs, languages, and creative domains as well as various sized models to consider performance restraints. This evolution will necessitate organizations to purse a more nuanced approach toward model selection, in which models are evaluated based on their strengths and limitations.
Elevate Your Business with AdEx Partners’ AI Expertise
Key Takeaways
AdEx Partners offers strategic support to evaluate and leverage Artificial Intelligence for your unique use cases
Our expertise in project management and technical knowledge helps assess the potential benefits and savings from AI for your business
We collaborate with you to create a tailored AI strategy that maximizes long-term value and remains relevant beyond current trends
At AdEx Partners, we support you in strategically evaluating and utilizing the potential of Artificial Intelligence for your specific use cases. Drawing on our extensive experience in project and portfolio management, combined with deep technical knowledge and industry insight, AdEx Partners can guide you through the emerging AI landscape, assess the potential benefits and savings for your business, and help manage your portfolio of AI use cases for optimal performance and innovation.
Our approach aims to help you utilizing the right technologies to maximize long-term business value. We work hand in hand with you to develop a customized AI strategy that is not only tailored to your specific requirements and needs, but that will also last consistently beyond passing trends such as the current ChatGPT hype.
“Ready to unlock the full potential of AI for your organization? Our looks forward to discussing your unique challenges and explore how we can support your journey toward AI-driven success. Don’t miss the opportunity to stay ahead of the curve – reach out to us for further information or to schedule a meeting!”
