When there's a lot of noise around a particular technology topic, it signals an opportunity to dig deeper and understand whether it deserves our attention and investment. Artificial intelligence has been the field that has recently captured my focus the most, and the focus of Brightgrove as a whole, too. 

Nearly a decade ago, I was a regular at AI meetups and conferences. Back then, we mostly referred to specific tools and elements — computer vision, machine and deep learning algorithms, genetic algorithms — rather than the broad umbrella term "AI." Recently, while reviewing literature on business applications of AI, I encountered the recurring issue of algorithmic bias, which prompted me to reflect more deeply on this subject. 

Below are my observations on the matter, informed by my experience and our team's work building software solutions with AI technologies. I hope these thoughts contribute to ongoing discussions in our industry. 

The Problem of Biased AI 

A clear example of biased AI emerged from a system designed to assist judges in making objective decisions. The court eventually discontinued its use after discovering that the system produced biased judgments against specific ethnic groups. 

The root cause? The training data was neither equally representative nor of uniform quality. In other words, biased data led to biased assessments. Since then, major tech companies have largely withdrawn from providing AI solutions in areas where algorithmic decisions directly impact human lives. 

This observation led me to a fundamental question: where does bias in AI originate? The data used to train AI models is collected and curated by humans, who are inherently subjective. 

The algorithms themselves are designed by humans, who may — intentionally or unintentionally — introduce bias into their work. It's a troubling reality: combining biased data with biased design does not yield objectivity. 

The Human Factor 

This line of reasoning brings me to a broader question: can we ever create a truly objective artificial intelligence, one that considers all possible solutions fairly and makes impartial judgments? 

Let's examine this systematically. The creators of AI are humans, and humans are biased by nature. Setting aside philosophical considerations, let's focus solely on the material aspects of human existence and how they shape decision-making. 

A newborn represents biological hardware with basic learning algorithms. This "hardware" is constructed through the genetic contribution of two parents. It's reasonable to conclude that genetic makeup influences behavioral tendencies and cognitive patterns. Therefore, even a newly formed human system arrives with inherited predispositions — a form of initial bias. 

Data and Development 

Another factor that shapes human subjectivity — and consequently, the AI systems we build — is the data we're exposed to throughout our development. 

Initially, an individual's training data comes from a limited, closed circle: parents, relatives, immediate community. As we mature and expand our social networks, the volume and variety of information increases, but it remains constrained by environmental factors. 

One might argue that before the internet and modern communication technologies, social groups were isolated, information didn't flow freely, and people operated with significant cognitive limitations. Today, we have near-instant access to vast information repositories and can connect with individuals worldwide. However, there are critical caveats: 

Algorithmic curation and moderation shape information access. Platforms and digital resources use algorithms to deliver "relevant" content — essentially creating a biased training set by default. Even organizational knowledge bases and documentation systems reflect the priorities and perspectives of their creators. 

Individual attention is inherently biased. Even with efforts to diversify our information sources, our consciousness, prior experience, and professional networks predispose us to notice and value certain information over others. We unintentionally seek and process data in subjective ways. 

Organizations can invest in broadening knowledge bases and fostering diverse perspectives to develop more objective decision-making processes. However, I've observed that both individuals and teams, after reaching a peak of objectivity, tend to narrow their focus as established patterns and preferences solidify. 

There's a practical reason for this: to maintain objectivity, we must actively engage with diverse knowledge domains, or risk losing access to alternative perspectives. The less frequently we consider certain viewpoints or methodologies, the less likely we are to incorporate them into our problem-solving approaches. 

As knowledge accumulates, more effort is required to maintain a balanced perspective. It seems that development teams and organizations must constantly work to counter increasing subjectivity as systems and processes mature. 

Professional Bias and Cross-Functional Communication 

One of the most significant sources of bias in software development and AI implementation stems from professional specialization. Each role within a development organization views problems through a distinct lens shaped by training, experience, and daily responsibilities. 

Role-Specific Perspectives: 

Engineers often gravitate toward technical elegance and implementation feasibility. When evaluating whether to use AI for a particular problem, developers may focus on the sophistication of the algorithms, the challenge of the implementation, or the opportunity to work with cutting-edge technologies. This can lead to over-engineering or applying AI where simpler solutions would suffice. I've seen this firsthand in our projects where the engineering team's enthusiasm for a novel approach sometimes needed to be balanced against practical delivery timelines. 

Software architects consider system-wide implications, scalability, and long-term maintainability. They may favor solutions that fit established patterns or align with existing infrastructure, potentially dismissing innovative AI approaches that require architectural changes, or conversely, pushing for AI implementations to modernize legacy systems regardless of actual need. 

Quality assurance specialists prioritize testability, reliability, and edge case coverage. When evaluating AI solutions, QA teams may focus heavily on the challenges of testing non-deterministic systems, potentially creating resistance to AI adoption even when it's appropriate, or may not adequately account for the unique testing requirements that AI systems demand. 

Product owners and managers view problems through the lens of business value, user needs, and market positioning. They may push for AI features because competitors are implementing them or because "AI-powered" has marketing appeal, rather than because the technology genuinely solves user problems better than alternatives. 

Designers focus on user experience, accessibility, and interaction patterns. They may resist AI implementations that create unpredictable or opaque user experiences, or alternatively, may envision AI capabilities that aren't technically feasible within project constraints. 

The Communication Challenge

Effective cross-functional communication is essential for making unbiased technical decisions, but it's complicated by several factors that I've witnessed throughout my career: 

Different vocabularies and mental models: Each discipline uses specialized terminology and conceptual frameworks. What an engineer means by "model accuracy" differs from what a product owner understands by that phrase. Bridging these semantic gaps requires conscious effort and often fails under time pressure. 

Unequal representation in decision-making: In many organizations, certain voices carry more weight than others. Engineering teams may dominate technical architecture discussions, while business stakeholders may override technical concerns in roadmap planning. This power imbalance means not all perspectives receive fair consideration. 

Confirmation bias in team dynamics: Teams tend to seek information that confirms their existing preferences. If a team has already decided that AI is the solution, subsequent discussions may focus on how to implement it rather than whether it's appropriate. Dissenting voices may be dismissed as obstructionist rather than constructively critical. 

The challenge of structured decision-making: Fair evaluation requires systematic comparison of alternatives with clearly defined criteria. However, many teams make decisions through informal consensus or defer to the highest-paid person's opinion. Without structured frameworks for weighing options, bias inevitably creeps in. 

Strategies for Reducing Professional Bias: Through my work with various teams, I've seen several approaches succeed in reducing professional bias: 

Cross-functional problem framing sessions: Before jumping to solutions, diverse teams collaborate to thoroughly understand and define the problem from multiple angles. This prevents premature convergence on a particular approach. 

Structured decision-making frameworks: Using methodologies like decision matrices, trade-off analysis, or even pre-mortems helps ensure that all options receive systematic evaluation against objective criteria. 

Rotating leadership in technical discussions: Allowing different roles to lead various phases of technical planning ensures that diverse perspectives shape the conversation from the start, not just respond to proposals. 

Devil's advocate assignments: Explicitly asking team members to argue against prevailing assumptions creates space for critical examination without personal conflict. 

Regular retrospectives on decision quality: Teams that review past decisions to understand which biases influenced outcomes can develop awareness and adjust their processes accordingly. 

Despite these strategies, achieving truly unbiased decision-making remains aspirational. The personalities involved, organizational politics, time pressures, and the inherent uncertainty of software development all contribute layers of subjectivity. 

Biased Use Cases and AI Hype 

I appreciate when literature on AI business applications acknowledges that organizations may not actually need AI for every problem. There's a reason why statistics indicate that approximately 85% of AI projects fail to deliver expected value — AI is frequently applied where it's unnecessary or inappropriate. 

This isn't accidental. The current technological landscape creates strong bias toward AI adoption. The constant messaging — "AI will solve your problems," "AI will transform your business," "AI can do anything" — makes it difficult not to default to AI when evaluating solutions. 

But what is AI, really? It's marketed as artificial intelligence, but it's fundamentally a collection of tools, sometimes powerful enough to approximate intelligent behavior within narrow domains. 

The crucial step is understanding what AI actually encompasses, educating ourselves and our clients about underlying technologies — the differences between deep learning and traditional machine learning, between rule-based systems and neural networks — and approaching problems with objectivity before selecting implementation approaches. 

The Path Forward 

In the Marvel Cinematic Universe, Ultron, an AI, analyzed humanity and concluded that humans are fundamentally self-destructive. While this is fiction, it raises an interesting question: if an AI system were to analyze all available human-generated data, what conclusions might it reach? The global dataset of human behavior and decision-making isn't exactly flattering. 

This brings me to my central thesis: perhaps we shouldn't position modern AI as the ultimate solution to our challenges. Instead, AI serves best as an augmentation tool, helping us access, process, and comprehend vast amounts of data more effectively. 

AI can optimize, structure, and present information — functioning like advanced Intelligent Data Analysis (IDA) tools. So while it may seem paradoxical, we're creating biased tools to help us make less biased decisions. Perhaps, through this iterative process, bias combined with awareness of bias can move us incrementally toward greater objectivity. 

Conclusion 

Creating truly unbiased AI seems impossible because both the data and the design originate from humans, who are inherently subjective. Our genetic makeup, developmental environment, professional training, and organizational context all contribute layers of bias that inevitably influence the systems we build. 

The AI tools we develop today should not be viewed as replacements for human judgment but as extensions of it, augmenting our capabilities rather than supplanting them. Ideally, these tools help us process information more effectively and make decisions that are incrementally less biased, even if perfect objectivity remains unattainable. 

As someone working in software development, I approach emerging AI technologies as tools to enhance my work and deliver better solutions to our clients, not as infallible technologies to be deployed uncritically. As this field continues to evolve rapidly, I'm committed to ongoing learning and welcome dialogue with others navigating the complex intersection of AI, human nature, and objectivity in software development.