- The Wheat from the Chaff
- Posts
- The Wheat from the Chaff #18
The Wheat from the Chaff #18
This week, we are going to talk about a very effective strategy: returning to basics. Through an example, we will demonstrate how current tools can support this movement and present the potential of artificial intelligence in accelerating the search for new antibiotics.
Jhonata Emerick
August 17, 2023
As part of my job, every week I read a lot of content about data, technology, and artificial intelligence (also known as AI). I filter them and send you the best ones, every Sunday.
Hello everyone đź‘‹ ,
This week, we are going to talk about a very effective strategy: returning to basics. Through an example, we will demonstrate how current tools can support this movement and present the potential of artificial intelligence in accelerating the search for new antibiotics.
It's time to separate THE WHEAT FROM THE CHAFF…
Less is More
I had an interesting conversation this week about how artificial intelligence can be applied in a specific sector. In scenarios where it is complex to separate the signal from the noise, or the wheat from the chaff, returning to basics is a very effective strategy. Therefore, it is important to highlight that data is essential in any artificial intelligence system, as it is the pillar for the functioning of these modern algorithms.
To simplify, AI learns from data, just as students learn from textbooks and exercises. AI analyzes these "books" and "exercises," looking for patterns and rules. In some cases, the data comes with labels, which function as answers, helping to understand the context. Data labeling is a process similar to teaching a friend to identify animals, but in this case, we are teaching a computer. We show various images, for example, of a dog or a cat, and inform the computer what each image represents. Over time, the computer learns to recognize and differentiate these categories by itself. This process can be conducted by humans or with partial assistance from algorithms, which are step-by-step instructions that the computer follows.
AI systems, whether generative or not, rely on data labeling. Generative systems, for example, that create music or art, need labeled examples to learn how to generate their creations. Without data labeling, these systems cannot learn and enhance their skills. Current technologies are purportedly capable of creating original content, however, the quality and relevance of the generated content still depend on the data used to train them.
I understand up to this point, Jhon, but where are you heading?
Although all this technology can generate new and unique content, the quality of various aspects of the results depends heavily on the quality of the training data. In sectors where this data is not publicly accessible, it creates barriers to entry. For example, in the financial sector, the absence of historical and behavioral data of a company can make credit analysis by a financial institution difficult, especially if this institution does not have a prior relationship with the company. Despite the existence of public data and companies that provide such information, the lack of this historical data (most of it labeled) can affect the accuracy and relevance of the analysis.
However, there's no use being "a cobbler's house with wooden skewers," so let's talk about the data. You may have heard of Open Banking. This initiative by the Central Bank aims to bring innovation to the financial system, foster competition, and enhance the offering of financial products and services. Therefore, one of its goals is to reduce the information asymmetry we mentioned, ensuring that customers have full control over how, when, and with whom they share their data. However, for financial institutions to share customers' personal and financial data with other institutions, the customer must provide prior authorization, known as consent.
By the end of March, the total number of consents reached 28.3 million, according to data from Finsiders in partnership with consulting firm Bip. On the other hand, in 2023, the number of accounts opened in financial institutions in Brazil surpassed the milestone of 1 billion, according to data from the Central Bank. Therefore, a simple approximation suggests that the total consents represent about 2.83% of the total number of accounts. This data indicates that obtaining consent is still a challenge for the sector, but it is expected that this situation will progressively improve with the initiatives of the Central Bank. An important observation is that these are not unique accounts for individuals and legal entities, and consent is also not unique. Therefore, this is an approximation that highlights the long journey ahead for this data to be effectively used.
This example serves to provoke reflection on where and how to apply this arsenal of AI-based tools. In the case we mentioned, the focus should be on finding alternative ways to understand this customer to make better decisions. By doing so, we begin to collect behavioral data that will be beneficial for everyone.
To illustrate, let's conduct a test: suppose a company that operates a children's socks e-commerce requested credit from institution XPTO. Using some of these modern tools, I performed some data cross-referencing, for example.
On the website where the company claimed to have the highest sales volume, I conducted a search using a tool that provided me with a list of brands in that category, each seller's payment methods, shipping costs, and prices. This way, I can already determine if the seller is offering a competitive price, as well as check if the offered payment methods are suitable or even if they are using techniques such as free shipping combined with a below-market price.
I noticed that the store is offering a 12x installment plan (higher than competitors), with a price below the average of other stores for the same product. Additionally, the shipping fee for a certain region is lower than what competitors charge.
Shortly after, I started analyzing your e-commerce website, which represents the second-largest source of revenue. Using another tool, I found that the total number of unique visitors has been gradually decreasing. Additionally, I noticed that the quantity of visits is significantly lower when compared to competitors. Another relevant piece of information is that a large portion of visits originates from paid media. Therefore, it becomes essential to analyze if the customer acquisition cost, compared to their Lifetime Value (LTV), is truly sustainable.
My provocation is: how can we employ these powerful generative Artificial Intelligence tools to solve challenges with the perspective of generating value for the entire involved chain? In the example we discussed (simple yet elucidating), technology acts as a pillar supporting the business strategy, as it has the potential to reduce entry barriers and create new opportunities for growth and revenue generation. The era of AI is just beginning! :)
Enhancing our humanity with AI
Researchers from MIT and McMaster University used an artificial intelligence algorithm to discover a new antibiotic capable of combating the Acinetobacter baumannii bacterium. This bacterium is commonly found in hospital environments, is resistant to many existing drugs, and can cause serious diseases such as pneumonia and meningitis.
The algorithm was trained with data from around 7,500 chemical compounds to identify those capable of inhibiting the growth of the bacteria. This allowed the algorithm to learn the chemical characteristics associated with growth inhibition.
These tests resulted in nine antibiotics, including one that ended up being extremely effective in eliminating the bacteria. This discovery reinforces the potential of artificial intelligence in accelerating the search for new antibiotics.
Interesting topics in Brazil:
Trending topics Global:
We would love your help sharing this website!
If you know someone that might like our posts, don’t be shy and share it with them! By the way, we appreciate this!
See you Sunday…
We always arrive in your inbox around 12:08 PM. Some email servers are stubborn and delayed... Others are even worse and throw us into spam and/or promotions. What's the logic behind this timing? ...None :)