A Framework for Ethical Research and Innovation

Abstract

In this contribution, we set out a framework for ethical research and innovation. Our framework draws upon recent scholarly work recommending the introduction of new models at the intersection of ethics, strategy, and science and technology studies to inform and explicate how the decisions of researchers can be considered ethical. Ethical research and innovation is construed in our framework as a dynamic process emerging from decisions of multiple stakeholders in innovation ecosystems prior to, during and after the execution of a research and innovation project. The framework can be used by different types of research organizations to implement governance models of ethical research and innovation.

Appendix A: An Illustration

The illustration presented in this appendix is modeled on the SPICE project, a project aimed at informing decisions for the development of climate geoengineering technologies (Stilgoe et al. 2013).

First Phase

In the first phase of inclusion, the strategists need to be defined. To this end, the delineation of the ecosystem of the project is assumed to consist of two research organizations and a research-funding agency as strategists.

Second Phase

During the second phase, the alternatives need to be elicited. We assume that the project is currently transitioning from “lab tests to a field trial,” as was the case with the SPICE project (Stilgoe et al. 2013, p. 1575). As their legitimacy, urgency and power have increased, the strategists now face not only the fierce opposition from nongovernmental organizations but also the need to respond to them (Mitchell et al. 1997). The alternatives are: (1) to continue with the field trial ($a_1$), (2) to go back to previous phases to conduct additional broader impact assessments and reengage with external stakeholders ($a_2$), (3) to put the project on hold until conditions for the field trial are propitious ($a_3$), or (4) to abort the project ($a_4$).

Once the alternatives have been elicited, the next step in this second phase is to define the criteria to be used to analyze the alternatives. The reflexivity criteria to be used are compliance ($c_1$), internal communications ($c_2$), external communications ($c_3$) and reviews ($c_4$). The anticipation criteria are environmental risks ($c_5$), social risks ($c_6$), economical impact ($c_7$), and political impact ($c_8$). It should be noted that the list of potential risks listed above as criteria under the dimension of anticipation is included here for illustration purposes and is not meant to be comprehensive. In general, this list will depend on the project at hand. In the case of the SPICE project, political, social and environmental risks were very salient, though technological risks in the area of geoengineering were also present. This is to be compared with other project types, such as those in the emerging Industry 4.0, where technological risks (e.g. in the area of cybersecurity) take center stage and would probably be included in the list of relevant criteria by stakeholders (Radanliev et al. 2020).

With the set of strategists, alternatives and criteria in place, the next step in this second phase is to elicit the weights of criteria for each strategist. To this end, we follow the procedure proposed by Thomas Saaty as part of the analytical hierarchy process (AHP), a widely used multicriteria decision analysis method (Saaty 1980). This method deploys the Saaty scale in order to measure the relative importance of one criterion over another.¹⁵ To elicit the weights of criteria, we first multiply every value in each row and then raise the result to the power of 1/m, where m is the number of criteria. The resulting value for each row is then divided by sum of the values of all rows and gives the normalized weight for each criterion.

Applying this method, the profile of strategist $s_1$ gives higher importance to compliance with existing norms, internal communications and the economic impact of the project, as shown in Table 5.

Table 5 The weights of criteria for the first strategist

The profile of strategist $s_2$ is shown in Table 6. As we can see, complying with existing norms, communicating with internal stakeholders more than with external stakeholders, and assessing the project economic impact are more important to the second strategist than anticipating environmental and social risks.

Table 6 The weights of criteria for the second strategist

The profile of strategist $s_3$ is shown in Table 7. Strategist $s_3$ corresponds to a research-funding agency from the public sector. The third strategist is more concerned with the incorporation of external stakeholders and with the political risks of the project.¹⁶

Table 7 The weights of criteria for the third strategist

Once the set of alternatives, the set of criteria, and the m-dimensional vectors of weights (one per strategist) have been elicited, the evaluation matrix of alternatives containing the values $v_{ji}$ that each alternative $a_j$ delivers under each criterion $c_i$ is generated. This matrix $E_{mn} = v_{ji}$ is shown in Table 8.¹⁷

Table 8 The evaluation matrix of alternatives

Third Phase

The third phase is implemented using a multicriteria decision analysis method. We illustrate this process using the TODIM method proposed by Gomes and Lima (1991).

Using the evaluation matrix of alternatives $E_{nm} = [v_{ji}]$, which is one of the outputs of the second phase of our methodology, the value function $\phi ^{k.i}$ of TODIM computes a pairwise comparison of the values $v_{hi}$ and $v_{ji}$ that the pair of alternatives $(a_h, a_j)$ deliver under criterion $c_i$ in the evaluation matrix of alternatives $E_{nm} = [v_{ji}]$. The value function $\phi ^{k,i}$ yields $m \times l$ matrices $\varPhi ^{k, i}$, the partial dominance matrices of alternatives containing the values $\phi ^{k,i}_{hj}$ representing the partial dominance of alternative $a_h$ over alternative $a_j$ under criterion $c_i$ for strategist $s_k$, with $1 \le i \le m$, $1 \le h, j \le n$, and $1 \le k\le l$.

The value function $\phi ^{k,i}$ is given by the following expression:

$$\begin{aligned} \phi ^{k,i}(a_h, a_j) = \left\{ \begin{array}{lclrlrl} \sqrt{{w^k_{i}}\ {(v_{hi}-v_{ji})}} & {if } \left( v_{hi}-v_{ji}>0\right) &&\\ 0 & if \left( v_{hi}-v_{ji}=0\right) &\\ - \sqrt{\frac{v_{ji}-v_{hi}}{w^k_{i}}} & if \left( v_{hi}-v_{ji}<0\right) & \end{array} \right. \end{aligned}$$

(1)

The profile of each strategist is brought to bear in (1) by the weight $w^k_i$ that each strategist $s_k$ attaches to each criterion $c_i$. Using the partial dominance matrices $\varPhi ^{k, i}$ for each criterion and strategist, the final dominance matrix of alternatives is computed using the function $\delta ^k$, with $1 \le i \le m$, $1 \le h, j \le n$, and $1 \le k\le l$. Each of the l final dominance matrices is computed using the following expression:

$$\begin{aligned} \delta ^k\left( a_h,a_j\right) = \sum _{i=1}^{m}\ \phi ^{k, i}_{hj} \end{aligned}$$

(2)

Equation 2 generates l matrices of dominance of alternatives $\varDelta ^{k}$, one for each strategist, containing the values $\delta ^k_{hj}$ representing the dominance of alternative $a_h$ over alternative $a_j$ for strategist $s_k$. Each one of these dominance matrices corresponds to the evaluation matrices of strategists (EMS$_k$), with $1 \le k\le l$, which is the output of the first step of the third phase of the methodology proposed in the “Operationalizing the Framework” section of our article. Finally, the global value that each alternative $a_h$ yields for strategist $s^k$, with $1 \le h, j \le n$, and $1 \le k\le l$ is given by expression (3):

$$\begin{aligned} \xi ^k_{h} = \sum _{j=1}^n \delta ^k_{h,j} \end{aligned}$$

(3)

In order to generate the rankings as part of the second step of the third phase of our methodology, $\xi ^k_{h}$, the global value that each alternative $a_h$ yields for strategist $s^k$, with $1 \le h \le n$, and $1 \le k\le l$, needs to be normalized as per expression (4):

$$\begin{aligned} \hat{\xi }^k_{h} = \frac{\xi ^k_{h} - min_{h=1}^n \xi ^k_{h}}{max_{h=1}^n \xi ^k_{h} - min_{h=1}^n \xi ^k_{h}} \end{aligned}$$

(4)

The normalized global value of each alternative given by Eq. 4 leads to the ranking of alternatives for each strategist $s^k$, which is the output of the second step of the third phase of the methodology set out in the “Operationalizing the Framework” section of our article.

Applying Eqs. (1) through (4), we obtain three evaluation matrices of strategists (EMS$_k$), with $1 \le k\le 3$, and three rankings of all four alternatives, as shown in Tables 9, 10 and 11.

Table 9 Evaluation matrix and raking of alternatives for first strategist

Table 10 Matrix of gains, global value and ranking for second strategist

Table 11 Matrix of gains, global value and ranking for third strategist

Different profiles of strategists lead to different evaluation matrices of strategists, which may lead to different rankings of alternatives for strategists. In the case of our illustration, the first and third strategist would prefer to continue with the execution of the project by postponing the field trial and investing more resources in the phases of reflexivity and anticipation. The second strategist, on the other hand, would prefer to put the project on hold. For all three strategists, even aborting the project is a better strategy than continuing on with the field trial, as originally planned.

The last step during the third phase of our methodology would be to generate a consensus strategy. If the plurality principle were to be applied,¹⁸ then the second strategy would be the consensus strategy to be pursued. The strategists may agree to apply other rules to arrive at a consensus strategy (Fishburn 1973; Munda 2004), such as the Borda count (Borda 1784; Condorcet 1785; McLean and Urken 1995), which would lead to a different consensus strategy.

It is important to note that the methodology shown in Fig. 6 is iterative and dynamic in that at any given point in time, the flow of control can go back to previous phases to revisit decisions that have been already made, such as the inclusion of new stakeholders, which would require the flow of control to go back to the first phase, or the addition of new criteria or the modification of the profiles of strategists, which would require the flow of control to go back to the second phase. In the same way, other types of alternatives can emerge and be considered by the strategists by backtracking to the second phase of the lifecycle.

Our methodology does not endorse a particular multicriteria group decision analysis method. Different multicriteria group decision analysis methods can be used interchangeably to implement the third phase of our methodology. The choice will always depend on the type of multicriteria decision analysis problem at hand. While the TODIM method implemented in this case may be of interest to model the biases of human decision-making, especially those that arise in the domain of losses under deep uncertainty, in many cases the deployment of more computationally tractable methods, such as the TOPSIS method (Hwang and Yoon 1981), may be preferred, especially when considering large sets of criteria, alternatives, and strategists.