AI – From Data to Business Decisions
Collecting data is no longer a major challenge. But how should it be used? And how can it be implemented in AI systems? It is when data is used to streamline operations or find new applications that opportunities arise—opportunities that are captured with a clear purpose, an open mind, and well-structured data.
More and more companies are starting to look at artificial intelligence and machine learning on a larger scale. In a study where the Info-Tech Research Group asked CIOs and other IT leaders worldwide about their AI plans for 2024, one in three said they would roll out generative AI capabilities in the coming year. Most plan to use AI to automate repetitive tasks and IT operations, while two out of five also plan to use AI to define their business strategy.
Three Levels of Maturity
Kevin Söderberg, Senior Data Scientist and AI Tech Lead at HiQ, recognizes these ambitions. According to him, there are typically three ways a company starts looking at AI systems:
“There are still companies jumping on the hype without a clear idea of what they want to achieve, but more commonly we see a certain level of maturity where the company has collected data and wants to explore how it can be used. The third type is companies that have a strategy for how they want to leverage their data to create added value,” he explains, continuing:
“We encounter all three levels of maturity. The commonality is that these companies do not yet have in-house AI expertise and therefore do not always know what to do with the collected data. Additionally, there are few specialists in the market, making it challenging to bring the right skills into the company.”
The Development Pace Requires Constant Skill Development
Kevin Söderberg works with Albin Sidås, Data Scientist at HiQ, on visual perception for autonomous vehicles in a future project at Toyota. Working with AI systems is complex and requires expertise in multiple areas. For example, the current project at Toyota involves semantic segmentation—finding structures in an image, object identification—identifying objects in the image, and object tracking—following the identified objects.
“As AI is very broad and development progresses rapidly, we need to work in parallel while keeping ourselves updated in our respective fields,” says Albin Sidås.
Right Data at the Right Time
Data that can be used in AI models ranges from unstructured data, such as images and sounds, and semi-structured data, such as event reports and HTML objects, to structured data typically found in spreadsheets, relational databases, and tabular data generated from sensors.
The type of data you use depends on the application. In visual perception, unstructured data in the form of images is naturally used, while AI systems for detecting fraud, for example, often use structured data.
“Many are unaware of the limitations of different types of data and how to use the right data at the right time,” says Albin Sidås.
“So far, each AI system has used only one type of data, but we now see the emergence of so-called multimodal models, where different types of input data are combined to create more robust models,” says Kevin Söderberg.
Data Quality in Focus
ÄvEven if a company has collected large amounts of data for a specific purpose, it is not certain that the data is sufficient. Albin Sidås describes challenges that can arise with recorded real-world data as an example.
“It can be difficult to get good quality and sufficient variation in recorded data. To solve this, you need to work with different lighting conditions, different types of cameras and lenses, and recording in various environments; otherwise, you risk ending up with a large amount of duplicated data that is of no use. If that’s not enough, you can supplement with synthetic data or use open datasets available, mainly from universities, to improve your models,” he says.
“It can be difficult to get good quality and sufficient variation in recorded data. To solve this, you need to work with different lighting conditions, different types of cameras and lenses, and recording in various environments; otherwise, you risk ending up with a large amount of duplicated data that is of no use.”
Albin Sidås
Lower the Barrier and Start Small
Both Kevin Söderberg and Albin Sidås find that Swedish companies are interested and fairly good at adopting AI and machine learning, but there is also some uncertainty about how AI systems can be utilized and whether the investment is worthwhile.
“A tip is to start with low-hanging fruit and demonstrate results through a Proof of Concept, such as using Large Language Models to create document summaries for efficiency. This usually helps organizations understand that AI systems are worth investing in,” says Kevin Söderberg.
AI Is Becoming More Common in Product Development
AI is currently mainly found in three areas: building intelligent products, building intelligent services, and improving a company’s internal processes, often focusing on products and services.
“A company’s technical maturity and ability to embrace change are crucial in determining whether it invests in AI or not. It is more common to invest in the technology within product and service development. Internal processes often come lower on the list, even though a more efficient internal process can ultimately accelerate innovation in the offerings,” says Kevin Söderberg.
Requirements to Explain AI-Based Decisions
AI systems have many advantages, but there are also some limitations and challenges. They will change the way many of us work, and many repetitive tasks are already handled by intelligent assistants. But with that comes new tasks.
For example, models must be updated and retrained with new data, and the system does not yet do that independently.
“We need to understand that artificial intelligence is not an exact science. It is a statistical model, and it requires a specific mindset,” says Kevin Söderberg.
In AI, we work with probability rather than causality, which many are accustomed to. Many view AI as a black box whose output cannot be explained. But there are methods to examine AI models, including XAI (eXplainable Artificial Intelligence), which is a framework for analyzing AI models and understanding how decisions are made in the model.
“It can be difficult to provide a mathematical proof for a result, but we can empirically show how the model works and how it makes decisions,” says Albin Sidås.
Additionally, there are requirements that certain results, such as credit ratings, must be explainable. Within the EU, the European Centre for Algorithmic Transparency aims to ensure transparency around algorithms and models.
“We need to understand that artificial intelligence is not an exact science. It is a statistical model, and it requires a specific mindset.”
Kevin Söderberg
Success Factors for Your AI Initiative
- Access to data and hardware for training models and analyzing large amounts of data is a fundamental requirement.
- A strategy for how the company will work with data and AI – for direction and ensuring responsible AI
- An adapted project process that considers that AI systems require a more circular working method than traditional software development.
- An openness within the organization to the changes that results from models and analyses may lead to.
Get in touch!
Get in touch!
Choose office or contact HiQ International in Stockholm if you are in doubt.