TL; DR - My experience as a recent entrant to EA recruitment suggests that processes overly focus on skills and fail to assess attitudes early on in the process. This is at odds with my professional experience as a business user of talent, where attitudes are central to effective recruitment. Attitudes are harder to instil than skills, an indicator of high performance, and vital for when life doesn't go according to plan. Filtering initial recruiting rounds by attitude will improve hiring decisions, and include more candidates that are new to EA and lack context, but possess excellent attitudes.  

Introduction 

Reflecting on the excellent Why I'm excited about AI safety talent development initiatives, I’m keen to support the supply and development of talent into AI Safety. After over a quarter of century in management consulting, I’m looking to pivot into AI Safety and in doing so have reflections on EA recruitment from my initial experiences. In my career to date I’ve built large teams, scaled capabilities, set up and run tech graduate and apprenticeship schemes, and participated in and chaired more colleague appraisal rounds than I can care to remember - likely 50+. As such I have quite a lot of experience in recruiting and then appraising the performance of those recruited.  

Whilst my observations are an initial take, and provocative to generate discussion, hopefully they may assist EA's maturity journey by improving matching the right talent to the right roles. This post may be more relevant for hiring managers and talent specialists, but hopefully of interest for all looking to move or find roles, such as myself. 

This post concerns the assessment of attitudes in the EA recruitment process. By using the term ‘attitudes’, I’m referring to the definition that ChatGPT summarised to assist my laziness / productivity: “Attitudes in the workplace are individuals’ consistent patterns of thoughts, feelings, and predispositions toward their job, colleagues, or organisation, which influence how they perceive situations and behave at work. They have the following elements: 

• Cognitive element – what an employee thinks or believes (e.g., “This organisation values fairness”).
• Affective element – how they feel (e.g., pride, frustration, motivation).
• Behavioral element – their tendency to act in certain ways (e.g., cooperating, resisting change, going the extra mile).

In practice, workplace attitudes matter because they shape engagement, performance, collaboration, and culture.”

Observations

My initial take on the EA recruitment experience is that there seems to be a significant focus on capabilities over attitude. Namely, candidate skill sets are appraised to a far greater degree than their values, behaviours and motivations - the key components of their attitude to the role - if these are selected for at all.  This assertion has arisen from the following observations and experiences. 

1 - There seems to be a lack of focus on attitude early on in the recruitment process, with an overwhelming focus on skills. Timed tests seem to be de rigueur as an initial filter. Whilst these hurdles should help identify those with relevant skills, I'm not clear if this quality gate is that important to come so early in the process.  

2 - Screening for attitude later on in the process can miss those with borderline skills that could be relatively easily attained in role, at the risk of excluding those with excellent attitudes which can be harder to source. 

 3 - My professional experience is that skills and capabilities - assuming sufficient ability - can be attained far more easily than attitudinal fit, and the skills the candidate displays is less a predictor of success than how and why they use them. 

4 - My - admittedly commercial - experience disproportionately favours the importance of behaviours and motivations in predicting high performance. Whilst clearly recruiting someone who can do the job is important, how they do the job, and what they do when they can't do it, or if it goes wrong, is significant. I've recruited colleagues who have the necessary skills, but how they cope when they screw up - which we all do - is vital. A good example is a junior I recruited with a focus on attitude who made a massive error the day before we submitted a £20m+ bid; a market moving and share price impacting risk for the client global bank. Their action to own the issue, think around the problem and then work tirelessly to recover the mislaid information and disprove data leakage saved what would have been a career limiting outcome for many, myself included. Behaviour which is hard to predict if focusing on skills and not attitude. 

 5 - Attitude is critical to individual success. The old adage that perspiration beats inspiration and consistency beats intensity supports this. All the high performers I’ve encountered put the hard work in, and can be counted upon to perform when things are going well, and when they’re not. Life is hard to predict, and recruiting talent for a role can be difficult given the challenge in forecasting exactly what they will be doing for the next n months. Indeed, the ability to predict work exactly suggests the role should be automated. EA - and for that matter pretty much most knowledge economy roles - need individuals who can think and respond to new and unforeseen challenges. The world is littered with highly talented people who didn't persist. Slightly less talented people who do make a massive impact. This attitude is hard to select for in timed work tests, unless the candidate gives up. In my early days running a tech graduate scheme, my greatest errors were in recruiting massively skilled individuals who then proved high maintenance and difficult when the going got rough, and we iterated our recruitment process to address this. 

 6 - This is not an argument for mediocre capabilities. Rather, it’s a recognition that capability acquisition is usually easier than attitudinal fit. Getting both is ideal, but filtering for the thing harder to source later on in the process would suggest the process is wasteful. Anecdotally most EA organisations are struggling with too many applicants, so aiming for both should be entirely feasible from such large and talented applicant pools. These numbers make it all the more confusing that attitudes are so often not part of the process, or a late-process addendum.

7 - Anecdotal feedback from connections in the EA community have highlighted staff issues arising from underperformance on attitude. A few individuals involved in talent management for EA with whom I’ve discussed this topic have volunteered challenges arising from behaviours, and the recognition that the cause of the problems was in the hiring process, which emphasised skills over attitude. 

 8 - Input from peers making similar career pivots into EA has observed the following:

• In my experience, the most difficult colleagues/employees are those who are unreliable (missing deadlines, needing frequent chasing, generally disorganized or inefficient), not those who lack skills. Skills are only valuable when they're correctly applied, after all. I've been really surprised to see these issues even from people working at the most elite of EA orgs, who must have been selected from incredibly large and competitive pools.
• ‘Soft skills’ or attitudes are just as important, if not more, to work effectively with colleagues. Maybe if they’d included paid real life work tests instead of the other work test types, to see how people perform in a professional environment, they’d update their hiring process into a more efficient one. After applying for 17 jobs, I’ve only seen CEA do this, and was, indeed, at the end of the process.
• I was surprised to see lack of adaptability, communication skills, unreliability, resistance to get/give candid feedback, lack of empathy… If there is interest in building community and bringing more professionals into the space, it would make sense to give feedback on work tests as not all of them are paid. It would be great to give the chance to the person applying to learn about what EA orgs expect, as solving a work test is something people can learn/improve, easier than soft skills.

9 - Further to the career pivot angle, this excellent forum post on Why experienced professionals fail to land high-impact roles helpfully highlights the importance of both EA context and skills. However, unless attitudes are assumed as being aligned with context, so greater EA context improves attitudinal alignment - which from the above experience I would assert that they aren’t always - attitudes are glaring by their absence. It could be that they don’t matter at all and have thus been omitted given the topic, or that ‘EA Context’ is a proxy for aligned values. However I would argue they’re still critical to anyone joining an EA organisation, experienced or not. The importance of traits / attitudes to EA is also highlighted in the Meta Coordination Forum (2024) / Talent Need Survey together with the lower importance of EA experience for hiring roles - admittedly from a very small sample size. This indicates that rather than relying on high EA context to assume alignment with EA attitude, directly selecting for attitudes as part of the recruitment process would help include more applicants new to EA or with lower context.  

Recommendations  

Whilst not a recruitment specialist, here are five high-level suggestions that can be used to tweak recruitment processes to promote the identification of attitudinal fit in candidates.

1. Define and publish organisational values at a level that can be evidenced for behavioural compliance. E.g. if learning and transparency are key values, behaviour that fails to accept responsibility and solve to prevent issue recurrence is clearly non-values aligned. Taking the top trait for the next five years from the Meta Coordination Forum (2024) / Talent Need Survey, judgement can only be deemed ‘good’ if it can be appraised against the wider values an organisation is seeking to promote.
2. Institute / bring forward assessment of attitude alignment to organisational values at the start of the recruitment process. These could be included in initial form-based and time-bound screening, requesting candidates to describe specific past situations and how they handled them, with simple scoring against organisational values. E.g. Explain a time you made a mistake, what happened etc. Successful candidates will be interviewed on submissions should they progress, to expand upon and verify. I’ve successfully used gamification tools to select for these as a fun but insightful approach to certain values, however solutions require investment, precluding small organisations and non-volume hiring. 
3. Re-orient capability / skills assessment to reflect organizational values. E.g. If integrity is a value, ensure the assessment has scope for candidates to reflect on their performance, identify what they found challenging and why, and what they would do differently if they were to take the assessment again. 
4. Interview successful candidates on their attitudinal submissions to expand and validate. Interrogate for additional scenarios they may encounter in the role, requiring them to explain prior experience and drill down into their approach and outcomes. Ensure some of these include foreseen but unwanted challenges - i.e. screw ups - to gauge attitudinal approach and fit. 
5. Componentise feedback to quickly and affordably categorise why candidates are not being progressed. Pro-forma responses explaining the relevant category can be provided to unsuccessful candidates at each process stage to help them improve their fit for future applications, benefiting the wider EA job space. E.g. Lack of evidence of skills / attitudes alignment / experience etc. 

Many EA organisations will be currently doing some of this, and this may be more relevant for more mature organisations. Regardless, I believe there is value in adoption. Of course, as a prior business user of recruitment outcomes rather than being a recruiter myself I may have gotten this entirely wrong. I don't have EA stats on application numbers per successful EA role. Over the years I’ve seen numerous benchmarking reports commissioned from boutique recruitment advisors placing spurious values on process elements, many of them disagreeing with each other, suggesting to me at least that the process can be more art than science. But what I have seen obliquely in EA is that the recruitment filtration processes appear to be at odds to those I've been accustomed to in professional services. Given discussion over Is EA still 'talent-constrained'? and the metrics over the number of applications required to land an EA role - 50+ anecdotally from my HIP IAP fellow experiences - there does seem some room to optimise. Or at least consider possibilities to do so. All thoughts and challenges are most welcome!